blob: d754e3cf99a85af0c433e71846db6ee051751ebd [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>
Al Viro719ea2f2013-09-29 11:24:49 -040029#include <linux/device.h>
David Howells07f3f052006-09-30 20:52:18 +020030#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Jens Axboed0bceac2009-05-18 08:20:32 +020032/*
Wu Fengguangbc31b862012-01-07 20:41:55 -060033 * 4MB minimal write chunk size
34 */
35#define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10))
36
37/*
Jens Axboec4a77a62009-09-16 15:18:25 +020038 * Passed into wb_writeback(), essentially a subset of writeback_control
39 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020040struct wb_writeback_work {
Jens Axboec4a77a62009-09-16 15:18:25 +020041 long nr_pages;
42 struct super_block *sb;
Jan Kara0dc83bd2014-02-21 11:19:04 +010043 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;
Dave Chinner7747bd42013-07-02 22:38:35 +100049 unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -060050 enum wb_reason reason; /* why was writeback initiated? */
Jens Axboec4a77a62009-09-16 15:18:25 +020051
Jens Axboe8010c3b2009-09-15 20:04:57 +020052 struct list_head list; /* pending work list */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020053 struct completion *done; /* set if the caller waits */
Jens Axboe03ba3782009-09-09 09:08:54 +020054};
55
Adrian Bunkf11b00f2008-04-29 00:58:56 -070056/**
57 * writeback_in_progress - determine whether there is writeback in progress
58 * @bdi: the device's backing_dev_info structure.
59 *
Jens Axboe03ba3782009-09-09 09:08:54 +020060 * Determine whether there is writeback waiting to be handled against a
61 * backing device.
Adrian Bunkf11b00f2008-04-29 00:58:56 -070062 */
63int writeback_in_progress(struct backing_dev_info *bdi)
64{
Jan Kara81d73a32010-08-11 14:17:44 -070065 return test_bit(BDI_writeback_running, &bdi->state);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070066}
Theodore Ts'o00d4e732012-09-19 22:42:36 -040067EXPORT_SYMBOL(writeback_in_progress);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070068
Jan Kara692ebd12010-09-21 11:51:01 +020069static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
70{
71 struct super_block *sb = inode->i_sb;
Jan Kara692ebd12010-09-21 11:51:01 +020072
Jan Karaa8855992013-07-09 22:36:45 +080073 if (sb_is_blkdev_sb(sb))
Christoph Hellwigaaead252010-10-04 14:25:33 +020074 return inode->i_mapping->backing_dev_info;
75
76 return sb->s_bdi;
Jan Kara692ebd12010-09-21 11:51:01 +020077}
78
Nick Piggin7ccf19a2010-10-21 11:49:30 +110079static inline struct inode *wb_inode(struct list_head *head)
80{
81 return list_entry(head, struct inode, i_wb_list);
82}
83
Wu Fengguang15eb77a2012-01-17 11:18:56 -060084/*
85 * Include the creation of the trace points after defining the
86 * wb_writeback_work structure and inline functions so that the definition
87 * remains local to this file.
88 */
89#define CREATE_TRACE_POINTS
90#include <trace/events/writeback.h>
91
Jan Kara65850272011-01-13 15:45:44 -080092static void bdi_queue_work(struct backing_dev_info *bdi,
93 struct wb_writeback_work *work)
94{
95 trace_writeback_queue(bdi, work);
96
97 spin_lock_bh(&bdi->wb_lock);
98 list_add_tail(&work->list, &bdi->work_list);
Artem Bityutskiy64677162010-07-25 14:29:22 +030099 spin_unlock_bh(&bdi->wb_lock);
Tejun Heo839a8e82013-04-01 19:08:06 -0700100
101 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102}
103
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200104static void
105__bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600106 bool range_cyclic, enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200108 struct wb_writeback_work *work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200109
Jens Axboebcddc3f2009-09-13 20:07:36 +0200110 /*
111 * This is WB_SYNC_NONE writeback, so if allocation fails just
112 * wakeup the thread for old dirty data writeback
113 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200114 work = kzalloc(sizeof(*work), GFP_ATOMIC);
115 if (!work) {
Tejun Heo839a8e82013-04-01 19:08:06 -0700116 trace_writeback_nowork(bdi);
117 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200118 return;
Jens Axboebcddc3f2009-09-13 20:07:36 +0200119 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200120
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200121 work->sync_mode = WB_SYNC_NONE;
122 work->nr_pages = nr_pages;
123 work->range_cyclic = range_cyclic;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600124 work->reason = reason;
Christoph Hellwigf0fad8a2009-09-11 09:47:56 +0200125
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200126 bdi_queue_work(bdi, work);
Jens Axboeb6e51312009-09-16 15:13:54 +0200127}
128
129/**
130 * bdi_start_writeback - start writeback
131 * @bdi: the backing device to write from
132 * @nr_pages: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +0800133 * @reason: reason why some writeback work was initiated
Jens Axboeb6e51312009-09-16 15:13:54 +0200134 *
135 * Description:
136 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300137 * started when this function returns, we make no guarantees on
Jens Axboe0e3c9a22010-06-01 11:08:43 +0200138 * completion. Caller need not hold sb s_umount semaphore.
Jens Axboeb6e51312009-09-16 15:13:54 +0200139 *
140 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600141void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
142 enum wb_reason reason)
Jens Axboeb6e51312009-09-16 15:13:54 +0200143{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600144 __bdi_start_writeback(bdi, nr_pages, true, reason);
Christoph Hellwigc5444192010-06-08 18:15:15 +0200145}
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800146
Christoph Hellwigc5444192010-06-08 18:15:15 +0200147/**
148 * bdi_start_background_writeback - start background writeback
149 * @bdi: the backing device to write from
150 *
151 * Description:
Jan Kara65850272011-01-13 15:45:44 -0800152 * This makes sure WB_SYNC_NONE background writeback happens. When
153 * this function returns, it is only guaranteed that for given BDI
154 * some IO is happening if we are over background dirty threshold.
155 * Caller need not hold sb s_umount semaphore.
Christoph Hellwigc5444192010-06-08 18:15:15 +0200156 */
157void bdi_start_background_writeback(struct backing_dev_info *bdi)
158{
Jan Kara65850272011-01-13 15:45:44 -0800159 /*
160 * We just wake up the flusher thread. It will perform background
161 * writeback as soon as there is no other work to do.
162 */
Wu Fengguang71927e82011-01-13 15:45:46 -0800163 trace_writeback_wake_background(bdi);
Tejun Heo839a8e82013-04-01 19:08:06 -0700164 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165}
166
167/*
Dave Chinnera66979a2011-03-22 22:23:41 +1100168 * Remove the inode from the writeback list it is on.
169 */
170void inode_wb_list_del(struct inode *inode)
171{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600172 struct backing_dev_info *bdi = inode_to_bdi(inode);
Dave Chinnera66979a2011-03-22 22:23:41 +1100173
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600174 spin_lock(&bdi->wb.list_lock);
175 list_del_init(&inode->i_wb_list);
176 spin_unlock(&bdi->wb.list_lock);
177}
Dave Chinnera66979a2011-03-22 22:23:41 +1100178
179/*
Andrew Morton6610a0b2007-10-16 23:30:32 -0700180 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
181 * furthest end of its superblock's dirty-inode list.
182 *
183 * Before stamping the inode's ->dirtied_when, we check to see whether it is
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200184 * already the most-recently-dirtied inode on the b_dirty list. If that is
Andrew Morton6610a0b2007-10-16 23:30:32 -0700185 * the case then the inode must have been redirtied while it was being written
186 * out and we don't reset its dirtied_when.
187 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600188static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
Andrew Morton6610a0b2007-10-16 23:30:32 -0700189{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600190 assert_spin_locked(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200191 if (!list_empty(&wb->b_dirty)) {
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200192 struct inode *tail;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700193
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100194 tail = wb_inode(wb->b_dirty.next);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200195 if (time_before(inode->dirtied_when, tail->dirtied_when))
Andrew Morton6610a0b2007-10-16 23:30:32 -0700196 inode->dirtied_when = jiffies;
197 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100198 list_move(&inode->i_wb_list, &wb->b_dirty);
Andrew Morton6610a0b2007-10-16 23:30:32 -0700199}
200
201/*
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200202 * requeue inode for re-scanning after bdi->b_io list is exhausted.
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700203 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600204static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700205{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600206 assert_spin_locked(&wb->list_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100207 list_move(&inode->i_wb_list, &wb->b_more_io);
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700208}
209
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700210static void inode_sync_complete(struct inode *inode)
211{
Jan Kara365b94ae2012-05-03 14:47:55 +0200212 inode->i_state &= ~I_SYNC;
Jan Kara4eff96d2012-11-26 16:29:51 -0800213 /* If inode is clean an unused, put it into LRU now... */
214 inode_add_lru(inode);
Jan Kara365b94ae2012-05-03 14:47:55 +0200215 /* Waiters must see I_SYNC cleared before being woken up */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700216 smp_mb();
217 wake_up_bit(&inode->i_state, __I_SYNC);
218}
219
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700220static bool inode_dirtied_after(struct inode *inode, unsigned long t)
221{
222 bool ret = time_after(inode->dirtied_when, t);
223#ifndef CONFIG_64BIT
224 /*
225 * For inodes being constantly redirtied, dirtied_when can get stuck.
226 * It _appears_ to be in the future, but is actually in distant past.
227 * This test is necessary to prevent such wrapped-around relative times
Jens Axboe5b0830c2009-09-23 19:37:09 +0200228 * from permanently stopping the whole bdi writeback.
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700229 */
230 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
231#endif
232 return ret;
233}
234
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700235/*
Wang Sheng-Hui0e2f2b22012-09-11 08:28:18 +0800236 * Move expired (dirtied before work->older_than_this) dirty inodes from
Jan Kara697e6fe2012-03-09 07:26:22 -0800237 * @delaying_queue to @dispatch_queue.
Fengguang Wu2c136572007-10-16 23:30:39 -0700238 */
Wu Fengguange84d0a42011-04-23 12:27:27 -0600239static int move_expired_inodes(struct list_head *delaying_queue,
Fengguang Wu2c136572007-10-16 23:30:39 -0700240 struct list_head *dispatch_queue,
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600241 struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700242{
Shaohua Li5c034492009-09-24 14:42:33 +0200243 LIST_HEAD(tmp);
244 struct list_head *pos, *node;
Jens Axboecf137302009-09-24 15:12:57 +0200245 struct super_block *sb = NULL;
Shaohua Li5c034492009-09-24 14:42:33 +0200246 struct inode *inode;
Jens Axboecf137302009-09-24 15:12:57 +0200247 int do_sb_sort = 0;
Wu Fengguange84d0a42011-04-23 12:27:27 -0600248 int moved = 0;
Shaohua Li5c034492009-09-24 14:42:33 +0200249
Fengguang Wu2c136572007-10-16 23:30:39 -0700250 while (!list_empty(delaying_queue)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100251 inode = wb_inode(delaying_queue->prev);
Jan Kara0dc83bd2014-02-21 11:19:04 +0100252 if (work->older_than_this &&
253 inode_dirtied_after(inode, *work->older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700254 break;
Jan Karaa8855992013-07-09 22:36:45 +0800255 list_move(&inode->i_wb_list, &tmp);
256 moved++;
257 if (sb_is_blkdev_sb(inode->i_sb))
258 continue;
Jens Axboecf137302009-09-24 15:12:57 +0200259 if (sb && sb != inode->i_sb)
260 do_sb_sort = 1;
261 sb = inode->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200262 }
263
Jens Axboecf137302009-09-24 15:12:57 +0200264 /* just one sb in list, splice to dispatch_queue and we're done */
265 if (!do_sb_sort) {
266 list_splice(&tmp, dispatch_queue);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600267 goto out;
Jens Axboecf137302009-09-24 15:12:57 +0200268 }
269
Shaohua Li5c034492009-09-24 14:42:33 +0200270 /* Move inodes from one superblock together */
271 while (!list_empty(&tmp)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100272 sb = wb_inode(tmp.prev)->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200273 list_for_each_prev_safe(pos, node, &tmp) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100274 inode = wb_inode(pos);
Shaohua Li5c034492009-09-24 14:42:33 +0200275 if (inode->i_sb == sb)
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100276 list_move(&inode->i_wb_list, dispatch_queue);
Shaohua Li5c034492009-09-24 14:42:33 +0200277 }
Fengguang Wu2c136572007-10-16 23:30:39 -0700278 }
Wu Fengguange84d0a42011-04-23 12:27:27 -0600279out:
280 return moved;
Fengguang Wu2c136572007-10-16 23:30:39 -0700281}
282
283/*
284 * Queue all expired dirty inodes for io, eldest first.
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700285 * Before
286 * newly dirtied b_dirty b_io b_more_io
287 * =============> gf edc BA
288 * After
289 * newly dirtied b_dirty b_io b_more_io
290 * =============> g fBAedc
291 * |
292 * +--> dequeue for IO
Fengguang Wu2c136572007-10-16 23:30:39 -0700293 */
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600294static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700295{
Wu Fengguange84d0a42011-04-23 12:27:27 -0600296 int moved;
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600297 assert_spin_locked(&wb->list_lock);
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700298 list_splice_init(&wb->b_more_io, &wb->b_io);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600299 moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
300 trace_writeback_queue_io(wb, work, moved);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200301}
302
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100303static int write_inode(struct inode *inode, struct writeback_control *wbc)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200304{
Tejun Heo9fb0a7d2013-01-11 13:06:37 -0800305 int ret;
306
307 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) {
308 trace_writeback_write_inode_start(inode, wbc);
309 ret = inode->i_sb->s_op->write_inode(inode, wbc);
310 trace_writeback_write_inode(inode, wbc);
311 return ret;
312 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200313 return 0;
Fengguang Wu2c136572007-10-16 23:30:39 -0700314}
315
316/*
Jan Kara169ebd92012-05-03 14:48:03 +0200317 * Wait for writeback on an inode to complete. Called with i_lock held.
318 * Caller must make sure inode cannot go away when we drop i_lock.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200319 */
Jan Kara169ebd92012-05-03 14:48:03 +0200320static void __inode_wait_for_writeback(struct inode *inode)
321 __releases(inode->i_lock)
322 __acquires(inode->i_lock)
Christoph Hellwig01c03192009-06-08 13:35:40 +0200323{
324 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
325 wait_queue_head_t *wqh;
326
327 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100328 while (inode->i_state & I_SYNC) {
329 spin_unlock(&inode->i_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200330 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
Dave Chinner250df6e2011-03-22 22:23:36 +1100331 spin_lock(&inode->i_lock);
Richard Kennedy58a9d3d82010-05-24 14:32:38 -0700332 }
Christoph Hellwig01c03192009-06-08 13:35:40 +0200333}
334
335/*
Jan Kara169ebd92012-05-03 14:48:03 +0200336 * Wait for writeback on an inode to complete. Caller must have inode pinned.
337 */
338void inode_wait_for_writeback(struct inode *inode)
339{
340 spin_lock(&inode->i_lock);
341 __inode_wait_for_writeback(inode);
342 spin_unlock(&inode->i_lock);
343}
344
345/*
346 * Sleep until I_SYNC is cleared. This function must be called with i_lock
347 * held and drops it. It is aimed for callers not holding any inode reference
348 * so once i_lock is dropped, inode can go away.
349 */
350static void inode_sleep_on_writeback(struct inode *inode)
351 __releases(inode->i_lock)
352{
353 DEFINE_WAIT(wait);
354 wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
355 int sleep;
356
357 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
358 sleep = inode->i_state & I_SYNC;
359 spin_unlock(&inode->i_lock);
360 if (sleep)
361 schedule();
362 finish_wait(wqh, &wait);
363}
364
365/*
Jan Karaccb26b52012-05-03 14:47:58 +0200366 * Find proper writeback list for the inode depending on its current state and
367 * possibly also change of its state while we were doing writeback. Here we
368 * handle things such as livelock prevention or fairness of writeback among
369 * inodes. This function can be called only by flusher thread - noone else
370 * processes all inodes in writeback lists and requeueing inodes behind flusher
371 * thread's back can have unexpected consequences.
372 */
373static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
374 struct writeback_control *wbc)
375{
376 if (inode->i_state & I_FREEING)
377 return;
378
379 /*
380 * Sync livelock prevention. Each inode is tagged and synced in one
381 * shot. If still dirty, it will be redirty_tail()'ed below. Update
382 * the dirty time to prevent enqueue and sync it again.
383 */
384 if ((inode->i_state & I_DIRTY) &&
385 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
386 inode->dirtied_when = jiffies;
387
Jan Kara4f8ad652012-05-03 14:48:00 +0200388 if (wbc->pages_skipped) {
389 /*
390 * writeback is not making progress due to locked
391 * buffers. Skip this inode for now.
392 */
393 redirty_tail(inode, wb);
394 return;
395 }
396
Jan Karaccb26b52012-05-03 14:47:58 +0200397 if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
398 /*
399 * We didn't write back all the pages. nfs_writepages()
400 * sometimes bales out without doing anything.
401 */
402 if (wbc->nr_to_write <= 0) {
403 /* Slice used up. Queue for next turn. */
404 requeue_io(inode, wb);
405 } else {
406 /*
407 * Writeback blocked by something other than
408 * congestion. Delay the inode for some time to
409 * avoid spinning on the CPU (100% iowait)
410 * retrying writeback of the dirty page/inode
411 * that cannot be performed immediately.
412 */
413 redirty_tail(inode, wb);
414 }
415 } else if (inode->i_state & I_DIRTY) {
416 /*
417 * Filesystems can dirty the inode during writeback operations,
418 * such as delayed allocation during submission or metadata
419 * updates after data IO completion.
420 */
421 redirty_tail(inode, wb);
422 } else {
423 /* The inode is clean. Remove from writeback lists. */
424 list_del_init(&inode->i_wb_list);
425 }
426}
427
428/*
Jan Kara4f8ad652012-05-03 14:48:00 +0200429 * Write out an inode and its dirty pages. Do not update the writeback list
430 * linkage. That is left to the caller. The caller is also responsible for
431 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 */
433static int
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700434__writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 struct address_space *mapping = inode->i_mapping;
Wu Fengguang251d6a42010-12-01 17:33:37 -0600437 long nr_to_write = wbc->nr_to_write;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200438 unsigned dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 int ret;
440
Jan Kara4f8ad652012-05-03 14:48:00 +0200441 WARN_ON(!(inode->i_state & I_SYNC));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442
Tejun Heo9fb0a7d2013-01-11 13:06:37 -0800443 trace_writeback_single_inode_start(inode, wbc, nr_to_write);
444
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 ret = do_writepages(mapping, wbc);
446
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100447 /*
448 * Make sure to wait on the data before writing out the metadata.
449 * This is important for filesystems that modify metadata on data
Dave Chinner7747bd42013-07-02 22:38:35 +1000450 * I/O completion. We don't do it for sync(2) writeback because it has a
451 * separate, external IO completion path and ->sync_fs for guaranteeing
452 * inode metadata is written back correctly.
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100453 */
Dave Chinner7747bd42013-07-02 22:38:35 +1000454 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) {
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100455 int err = filemap_fdatawait(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 if (ret == 0)
457 ret = err;
458 }
459
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400460 /*
461 * Some filesystems may redirty the inode during the writeback
462 * due to delalloc, clear dirty metadata flags right before
463 * write_inode()
464 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100465 spin_lock(&inode->i_lock);
Jan Kara6290be12012-05-03 14:47:57 +0200466 /* Clear I_DIRTY_PAGES if we've written out all dirty pages */
467 if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
468 inode->i_state &= ~I_DIRTY_PAGES;
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400469 dirty = inode->i_state & I_DIRTY;
470 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100471 spin_unlock(&inode->i_lock);
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100472 /* Don't write the inode if only I_DIRTY_PAGES was set */
473 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100474 int err = write_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 if (ret == 0)
476 ret = err;
477 }
Jan Kara4f8ad652012-05-03 14:48:00 +0200478 trace_writeback_single_inode(inode, wbc, nr_to_write);
479 return ret;
480}
481
482/*
483 * Write out an inode's dirty pages. Either the caller has an active reference
484 * on the inode or the inode has I_WILL_FREE set.
485 *
486 * This function is designed to be called for writing back one inode which
487 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
488 * and does more profound writeback list handling in writeback_sb_inodes().
489 */
490static int
491writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
492 struct writeback_control *wbc)
493{
494 int ret = 0;
495
496 spin_lock(&inode->i_lock);
497 if (!atomic_read(&inode->i_count))
498 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
499 else
500 WARN_ON(inode->i_state & I_WILL_FREE);
501
502 if (inode->i_state & I_SYNC) {
503 if (wbc->sync_mode != WB_SYNC_ALL)
504 goto out;
505 /*
Jan Kara169ebd92012-05-03 14:48:03 +0200506 * It's a data-integrity sync. We must wait. Since callers hold
507 * inode reference or inode has I_WILL_FREE set, it cannot go
508 * away under us.
Jan Kara4f8ad652012-05-03 14:48:00 +0200509 */
Jan Kara169ebd92012-05-03 14:48:03 +0200510 __inode_wait_for_writeback(inode);
Jan Kara4f8ad652012-05-03 14:48:00 +0200511 }
512 WARN_ON(inode->i_state & I_SYNC);
513 /*
Jan Karaf9b0e052013-12-14 04:21:26 +0800514 * Skip inode if it is clean and we have no outstanding writeback in
515 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
516 * function since flusher thread may be doing for example sync in
517 * parallel and if we move the inode, it could get skipped. So here we
518 * make sure inode is on some writeback list and leave it there unless
519 * we have completely cleaned the inode.
Jan Kara4f8ad652012-05-03 14:48:00 +0200520 */
Jan Karaf9b0e052013-12-14 04:21:26 +0800521 if (!(inode->i_state & I_DIRTY) &&
522 (wbc->sync_mode != WB_SYNC_ALL ||
523 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)))
Jan Kara4f8ad652012-05-03 14:48:00 +0200524 goto out;
525 inode->i_state |= I_SYNC;
526 spin_unlock(&inode->i_lock);
527
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700528 ret = __writeback_single_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600530 spin_lock(&wb->list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100531 spin_lock(&inode->i_lock);
Jan Kara4f8ad652012-05-03 14:48:00 +0200532 /*
533 * If inode is clean, remove it from writeback lists. Otherwise don't
534 * touch it. See comment above for explanation.
535 */
536 if (!(inode->i_state & I_DIRTY))
537 list_del_init(&inode->i_wb_list);
538 spin_unlock(&wb->list_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700539 inode_sync_complete(inode);
Jan Kara4f8ad652012-05-03 14:48:00 +0200540out:
541 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 return ret;
543}
544
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600545static long writeback_chunk_size(struct backing_dev_info *bdi,
546 struct wb_writeback_work *work)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600547{
548 long pages;
549
550 /*
551 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
552 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
553 * here avoids calling into writeback_inodes_wb() more than once.
554 *
555 * The intended call sequence for WB_SYNC_ALL writeback is:
556 *
557 * wb_writeback()
558 * writeback_sb_inodes() <== called only once
559 * write_cache_pages() <== called once for each inode
560 * (quickly) tag currently dirty pages
561 * (maybe slowly) sync all tagged pages
562 */
563 if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
564 pages = LONG_MAX;
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600565 else {
566 pages = min(bdi->avg_write_bandwidth / 2,
567 global_dirty_limit / DIRTY_SCOPE);
568 pages = min(pages, work->nr_pages);
569 pages = round_down(pages + MIN_WRITEBACK_PAGES,
570 MIN_WRITEBACK_PAGES);
571 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600572
573 return pages;
574}
575
Jens Axboe03ba3782009-09-09 09:08:54 +0200576/*
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800577 * Write a portion of b_io inodes which belong to @sb.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200578 *
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600579 * Return the number of pages and/or inodes written.
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800580 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600581static long writeback_sb_inodes(struct super_block *sb,
582 struct bdi_writeback *wb,
583 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200584{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600585 struct writeback_control wbc = {
586 .sync_mode = work->sync_mode,
587 .tagged_writepages = work->tagged_writepages,
588 .for_kupdate = work->for_kupdate,
589 .for_background = work->for_background,
Dave Chinner7747bd42013-07-02 22:38:35 +1000590 .for_sync = work->for_sync,
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600591 .range_cyclic = work->range_cyclic,
592 .range_start = 0,
593 .range_end = LLONG_MAX,
594 };
595 unsigned long start_time = jiffies;
596 long write_chunk;
597 long wrote = 0; /* count both pages and inodes */
598
Jens Axboe03ba3782009-09-09 09:08:54 +0200599 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100600 struct inode *inode = wb_inode(wb->b_io.prev);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200601
602 if (inode->i_sb != sb) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600603 if (work->sb) {
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200604 /*
605 * We only want to write back data for this
606 * superblock, move all inodes not belonging
607 * to it back onto the dirty list.
608 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600609 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200610 continue;
611 }
612
613 /*
614 * The inode belongs to a different superblock.
615 * Bounce back to the caller to unpin this and
616 * pin the next superblock.
617 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600618 break;
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200619 }
620
Christoph Hellwig9843b762010-10-24 19:40:46 +0200621 /*
Wanpeng Li331cbde2012-06-09 11:10:55 +0800622 * Don't bother with new inodes or inodes being freed, first
623 * kind does not need periodic writeout yet, and for the latter
Christoph Hellwig9843b762010-10-24 19:40:46 +0200624 * kind writeout is handled by the freer.
625 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100626 spin_lock(&inode->i_lock);
Christoph Hellwig9843b762010-10-24 19:40:46 +0200627 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100628 spin_unlock(&inode->i_lock);
Wu Fengguangfcc5c222011-07-11 23:08:50 -0700629 redirty_tail(inode, wb);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700630 continue;
631 }
Jan Karacc1676d2012-05-03 14:47:56 +0200632 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
633 /*
634 * If this inode is locked for writeback and we are not
635 * doing writeback-for-data-integrity, move it to
636 * b_more_io so that writeback can proceed with the
637 * other inodes on s_io.
638 *
639 * We'll have another go at writing back this inode
640 * when we completed a full scan of b_io.
641 */
642 spin_unlock(&inode->i_lock);
643 requeue_io(inode, wb);
644 trace_writeback_sb_inodes_requeue(inode);
645 continue;
646 }
Jan Karaf0d07b72012-05-03 14:47:59 +0200647 spin_unlock(&wb->list_lock);
648
Jan Kara4f8ad652012-05-03 14:48:00 +0200649 /*
650 * We already requeued the inode if it had I_SYNC set and we
651 * are doing WB_SYNC_NONE writeback. So this catches only the
652 * WB_SYNC_ALL case.
653 */
Jan Kara169ebd92012-05-03 14:48:03 +0200654 if (inode->i_state & I_SYNC) {
655 /* Wait for I_SYNC. This function drops i_lock... */
656 inode_sleep_on_writeback(inode);
657 /* Inode may be gone, start again */
Jan Karaead188f2012-06-08 17:07:36 +0200658 spin_lock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200659 continue;
660 }
Jan Kara4f8ad652012-05-03 14:48:00 +0200661 inode->i_state |= I_SYNC;
662 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200663
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600664 write_chunk = writeback_chunk_size(wb->bdi, work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600665 wbc.nr_to_write = write_chunk;
666 wbc.pages_skipped = 0;
Dave Chinner250df6e2011-03-22 22:23:36 +1100667
Jan Kara169ebd92012-05-03 14:48:03 +0200668 /*
669 * We use I_SYNC to pin the inode in memory. While it is set
670 * evict_inode() will wait so the inode cannot be freed.
671 */
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700672 __writeback_single_inode(inode, &wbc);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600673
674 work->nr_pages -= write_chunk - wbc.nr_to_write;
675 wrote += write_chunk - wbc.nr_to_write;
Jan Kara4f8ad652012-05-03 14:48:00 +0200676 spin_lock(&wb->list_lock);
677 spin_lock(&inode->i_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600678 if (!(inode->i_state & I_DIRTY))
679 wrote++;
Jan Kara4f8ad652012-05-03 14:48:00 +0200680 requeue_inode(inode, wb, &wbc);
681 inode_sync_complete(inode);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100682 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200683 cond_resched_lock(&wb->list_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600684 /*
685 * bail out to wb_writeback() often enough to check
686 * background threshold and other termination conditions.
687 */
688 if (wrote) {
689 if (time_is_before_jiffies(start_time + HZ / 10UL))
690 break;
691 if (work->nr_pages <= 0)
692 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800693 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600695 return wrote;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800696}
Nick Piggin38f21972009-01-06 14:40:25 -0800697
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600698static long __writeback_inodes_wb(struct bdi_writeback *wb,
699 struct wb_writeback_work *work)
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800700{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600701 unsigned long start_time = jiffies;
702 long wrote = 0;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800703
704 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100705 struct inode *inode = wb_inode(wb->b_io.prev);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800706 struct super_block *sb = inode->i_sb;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800707
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000708 if (!grab_super_passive(sb)) {
Wu Fengguang0e9958162011-07-29 22:14:35 -0600709 /*
710 * grab_super_passive() may fail consistently due to
711 * s_umount being grabbed by someone else. Don't use
712 * requeue_io() to avoid busy retrying the inode/sb.
713 */
714 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200715 continue;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800716 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600717 wrote += writeback_sb_inodes(sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200718 drop_super(sb);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800719
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600720 /* refer to the same tests at the end of writeback_sb_inodes */
721 if (wrote) {
722 if (time_is_before_jiffies(start_time + HZ / 10UL))
723 break;
724 if (work->nr_pages <= 0)
725 break;
726 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800727 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200728 /* Leave any unwritten inodes on b_io */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600729 return wrote;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200730}
731
Wanpeng Li7d9f0732013-09-11 14:22:40 -0700732static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600733 enum wb_reason reason)
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200734{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600735 struct wb_writeback_work work = {
736 .nr_pages = nr_pages,
737 .sync_mode = WB_SYNC_NONE,
738 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600739 .reason = reason,
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600740 };
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200741
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600742 spin_lock(&wb->list_lock);
Wu Fengguang424b3512010-07-21 20:11:53 -0600743 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600744 queue_io(wb, &work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600745 __writeback_inodes_wb(wb, &work);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600746 spin_unlock(&wb->list_lock);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200747
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600748 return nr_pages - work.nr_pages;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200749}
750
Wu Fengguangb00949a2010-11-18 14:38:33 -0600751static bool over_bground_thresh(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200752{
753 unsigned long background_thresh, dirty_thresh;
754
Wu Fengguang16c40422010-08-11 14:17:39 -0700755 global_dirty_limits(&background_thresh, &dirty_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200756
Wu Fengguangb00949a2010-11-18 14:38:33 -0600757 if (global_page_state(NR_FILE_DIRTY) +
758 global_page_state(NR_UNSTABLE_NFS) > background_thresh)
759 return true;
760
761 if (bdi_stat(bdi, BDI_RECLAIMABLE) >
762 bdi_dirty_limit(bdi, background_thresh))
763 return true;
764
765 return false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200766}
767
768/*
Wu Fengguange98be2d2010-08-29 11:22:30 -0600769 * Called under wb->list_lock. If there are multiple wb per bdi,
770 * only the flusher working on the first wb should do it.
771 */
772static void wb_update_bandwidth(struct bdi_writeback *wb,
773 unsigned long start_time)
774{
Wu Fengguangaf6a3112011-10-03 20:46:17 -0600775 __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
Wu Fengguange98be2d2010-08-29 11:22:30 -0600776}
777
778/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200779 * Explicit flushing or periodic writeback of "old" data.
780 *
781 * Define "old": the first time one of an inode's pages is dirtied, we mark the
782 * dirtying-time in the inode's address_space. So this periodic writeback code
783 * just walks the superblock inode list, writing back any inodes which are
784 * older than a specific point in time.
785 *
786 * Try to run once per dirty_writeback_interval. But if a writeback event
787 * takes longer than a dirty_writeback_interval interval, then leave a
788 * one-second gap.
789 *
790 * older_than_this takes precedence over nr_to_write. So we'll only write back
791 * all dirty pages if they are all attached to "old" mappings.
792 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200793static long wb_writeback(struct bdi_writeback *wb,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200794 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200795{
Wu Fengguange98be2d2010-08-29 11:22:30 -0600796 unsigned long wb_start = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600797 long nr_pages = work->nr_pages;
Jan Kara0dc83bd2014-02-21 11:19:04 +0100798 unsigned long oldest_jif;
Jan Karaa5989bd2009-09-16 19:22:48 +0200799 struct inode *inode;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600800 long progress;
Jens Axboe03ba3782009-09-09 09:08:54 +0200801
Jan Kara0dc83bd2014-02-21 11:19:04 +0100802 oldest_jif = jiffies;
803 work->older_than_this = &oldest_jif;
Jens Axboe03ba3782009-09-09 09:08:54 +0200804
Wu Fengguange8dfc302011-04-21 12:06:32 -0600805 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200806 for (;;) {
807 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800808 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200809 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200810 if (work->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200811 break;
812
813 /*
Jan Karaaa373cf2011-01-13 15:45:47 -0800814 * Background writeout and kupdate-style writeback may
815 * run forever. Stop them if there is other work to do
816 * so that e.g. sync can proceed. They'll be restarted
817 * after the other works are all done.
818 */
819 if ((work->for_background || work->for_kupdate) &&
820 !list_empty(&wb->bdi->work_list))
821 break;
822
823 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800824 * For background writeout, stop when we are below the
825 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200826 */
Wu Fengguangb00949a2010-11-18 14:38:33 -0600827 if (work->for_background && !over_bground_thresh(wb->bdi))
Jens Axboe03ba3782009-09-09 09:08:54 +0200828 break;
829
Jan Kara1bc36b62011-10-19 11:44:41 +0200830 /*
831 * Kupdate and background works are special and we want to
832 * include all inodes that need writing. Livelock avoidance is
833 * handled by these works yielding to any other work so we are
834 * safe.
835 */
Wu Fengguangba9aa832010-07-21 20:32:30 -0600836 if (work->for_kupdate) {
Jan Kara0dc83bd2014-02-21 11:19:04 +0100837 oldest_jif = jiffies -
Wu Fengguangba9aa832010-07-21 20:32:30 -0600838 msecs_to_jiffies(dirty_expire_interval * 10);
Jan Kara1bc36b62011-10-19 11:44:41 +0200839 } else if (work->for_background)
Jan Kara0dc83bd2014-02-21 11:19:04 +0100840 oldest_jif = jiffies;
Dave Chinner028c2dd2010-07-07 13:24:07 +1000841
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600842 trace_writeback_start(wb->bdi, work);
Wu Fengguange8dfc302011-04-21 12:06:32 -0600843 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600844 queue_io(wb, work);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200845 if (work->sb)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600846 progress = writeback_sb_inodes(work->sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200847 else
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600848 progress = __writeback_inodes_wb(wb, work);
849 trace_writeback_written(wb->bdi, work);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000850
Wu Fengguange98be2d2010-08-29 11:22:30 -0600851 wb_update_bandwidth(wb, wb_start);
Jens Axboe03ba3782009-09-09 09:08:54 +0200852
853 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200854 * Did we write something? Try for more
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600855 *
856 * Dirty inodes are moved to b_io for writeback in batches.
857 * The completion of the current batch does not necessarily
858 * mean the overall work is done. So we keep looping as long
859 * as made some progress on cleaning pages or inodes.
Jens Axboe71fd05a2009-09-23 19:32:26 +0200860 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600861 if (progress)
Jens Axboe71fd05a2009-09-23 19:32:26 +0200862 continue;
863 /*
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600864 * No more inodes for IO, bail
Jens Axboe71fd05a2009-09-23 19:32:26 +0200865 */
Wu Fengguangb7a24412010-07-21 22:19:51 -0600866 if (list_empty(&wb->b_more_io))
Jens Axboe03ba3782009-09-09 09:08:54 +0200867 break;
868 /*
Jens Axboe8010c3b2009-09-15 20:04:57 +0200869 * Nothing written. Wait for some inode to
870 * become available for writeback. Otherwise
871 * we'll just busyloop.
Jens Axboe03ba3782009-09-09 09:08:54 +0200872 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200873 if (!list_empty(&wb->b_more_io)) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600874 trace_writeback_wait(wb->bdi, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200875 inode = wb_inode(wb->b_more_io.prev);
Dave Chinner250df6e2011-03-22 22:23:36 +1100876 spin_lock(&inode->i_lock);
Jan Karaf0d07b72012-05-03 14:47:59 +0200877 spin_unlock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200878 /* This function drops i_lock... */
879 inode_sleep_on_writeback(inode);
Jan Karaf0d07b72012-05-03 14:47:59 +0200880 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200881 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200882 }
Wu Fengguange8dfc302011-04-21 12:06:32 -0600883 spin_unlock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200884
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600885 return nr_pages - work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200886}
887
888/*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200889 * Return the next wb_writeback_work struct that hasn't been processed yet.
Jens Axboe03ba3782009-09-09 09:08:54 +0200890 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200891static struct wb_writeback_work *
Minchan Kim08852b62010-08-03 12:51:16 +0200892get_next_work_item(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200893{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200894 struct wb_writeback_work *work = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200895
Artem Bityutskiy64677162010-07-25 14:29:22 +0300896 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200897 if (!list_empty(&bdi->work_list)) {
898 work = list_entry(bdi->work_list.next,
899 struct wb_writeback_work, list);
900 list_del_init(&work->list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200901 }
Artem Bityutskiy64677162010-07-25 14:29:22 +0300902 spin_unlock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200903 return work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200904}
905
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700906/*
907 * Add in the number of potentially dirty inodes, because each inode
908 * write can dirty pagecache in the underlying blockdev.
909 */
910static unsigned long get_nr_dirty_pages(void)
911{
912 return global_page_state(NR_FILE_DIRTY) +
913 global_page_state(NR_UNSTABLE_NFS) +
914 get_nr_dirty_inodes();
915}
916
Jan Kara65850272011-01-13 15:45:44 -0800917static long wb_check_background_flush(struct bdi_writeback *wb)
918{
Wu Fengguangb00949a2010-11-18 14:38:33 -0600919 if (over_bground_thresh(wb->bdi)) {
Jan Kara65850272011-01-13 15:45:44 -0800920
921 struct wb_writeback_work work = {
922 .nr_pages = LONG_MAX,
923 .sync_mode = WB_SYNC_NONE,
924 .for_background = 1,
925 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600926 .reason = WB_REASON_BACKGROUND,
Jan Kara65850272011-01-13 15:45:44 -0800927 };
928
929 return wb_writeback(wb, &work);
930 }
931
932 return 0;
933}
934
Jens Axboe03ba3782009-09-09 09:08:54 +0200935static long wb_check_old_data_flush(struct bdi_writeback *wb)
936{
937 unsigned long expired;
938 long nr_pages;
939
Jens Axboe69b62d02010-05-17 12:51:03 +0200940 /*
941 * When set to zero, disable periodic writeback
942 */
943 if (!dirty_writeback_interval)
944 return 0;
945
Jens Axboe03ba3782009-09-09 09:08:54 +0200946 expired = wb->last_old_flush +
947 msecs_to_jiffies(dirty_writeback_interval * 10);
948 if (time_before(jiffies, expired))
949 return 0;
950
951 wb->last_old_flush = jiffies;
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700952 nr_pages = get_nr_dirty_pages();
Jens Axboe03ba3782009-09-09 09:08:54 +0200953
Jens Axboec4a77a62009-09-16 15:18:25 +0200954 if (nr_pages) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200955 struct wb_writeback_work work = {
Jens Axboec4a77a62009-09-16 15:18:25 +0200956 .nr_pages = nr_pages,
957 .sync_mode = WB_SYNC_NONE,
958 .for_kupdate = 1,
959 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600960 .reason = WB_REASON_PERIODIC,
Jens Axboec4a77a62009-09-16 15:18:25 +0200961 };
962
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200963 return wb_writeback(wb, &work);
Jens Axboec4a77a62009-09-16 15:18:25 +0200964 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200965
966 return 0;
967}
968
969/*
970 * Retrieve work items and do the writeback they describe
971 */
Wanpeng Li25d130b2013-07-08 16:00:14 -0700972static long wb_do_writeback(struct bdi_writeback *wb)
Jens Axboe03ba3782009-09-09 09:08:54 +0200973{
974 struct backing_dev_info *bdi = wb->bdi;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200975 struct wb_writeback_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +0200976 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200977
Jan Kara81d73a32010-08-11 14:17:44 -0700978 set_bit(BDI_writeback_running, &wb->bdi->state);
Minchan Kim08852b62010-08-03 12:51:16 +0200979 while ((work = get_next_work_item(bdi)) != NULL) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200980
Dave Chinner455b2862010-07-07 13:24:06 +1000981 trace_writeback_exec(bdi, work);
982
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200983 wrote += wb_writeback(wb, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200984
985 /*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200986 * Notify the caller of completion if this is a synchronous
987 * work item, otherwise just free it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200988 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200989 if (work->done)
990 complete(work->done);
991 else
992 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200993 }
994
995 /*
996 * Check for periodic writeback, kupdated() style
997 */
998 wrote += wb_check_old_data_flush(wb);
Jan Kara65850272011-01-13 15:45:44 -0800999 wrote += wb_check_background_flush(wb);
Jan Kara81d73a32010-08-11 14:17:44 -07001000 clear_bit(BDI_writeback_running, &wb->bdi->state);
Jens Axboe03ba3782009-09-09 09:08:54 +02001001
1002 return wrote;
1003}
1004
1005/*
1006 * Handle writeback of dirty data for the device backed by this bdi. Also
Tejun Heo839a8e82013-04-01 19:08:06 -07001007 * reschedules periodically and does kupdated style flushing.
Jens Axboe03ba3782009-09-09 09:08:54 +02001008 */
Tejun Heo839a8e82013-04-01 19:08:06 -07001009void bdi_writeback_workfn(struct work_struct *work)
Jens Axboe03ba3782009-09-09 09:08:54 +02001010{
Tejun Heo839a8e82013-04-01 19:08:06 -07001011 struct bdi_writeback *wb = container_of(to_delayed_work(work),
1012 struct bdi_writeback, dwork);
Christoph Hellwig08243902010-06-19 23:08:22 +02001013 struct backing_dev_info *bdi = wb->bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +02001014 long pages_written;
1015
Tejun Heoef3b1012013-04-30 15:27:24 -07001016 set_worker_desc("flush-%s", dev_name(bdi->dev));
Peter Zijlstra766f9162010-10-26 14:22:45 -07001017 current->flags |= PF_SWAPWRITE;
Christoph Hellwig08243902010-06-19 23:08:22 +02001018
Tejun Heo839a8e82013-04-01 19:08:06 -07001019 if (likely(!current_is_workqueue_rescuer() ||
1020 list_empty(&bdi->bdi_list))) {
Artem Bityutskiy64677162010-07-25 14:29:22 +03001021 /*
Tejun Heo839a8e82013-04-01 19:08:06 -07001022 * The normal path. Keep writing back @bdi until its
1023 * work_list is empty. Note that this path is also taken
1024 * if @bdi is shutting down even when we're running off the
1025 * rescuer as work_list needs to be drained.
Artem Bityutskiy64677162010-07-25 14:29:22 +03001026 */
Tejun Heo839a8e82013-04-01 19:08:06 -07001027 do {
Wanpeng Li25d130b2013-07-08 16:00:14 -07001028 pages_written = wb_do_writeback(wb);
Tejun Heo839a8e82013-04-01 19:08:06 -07001029 trace_writeback_pages_written(pages_written);
1030 } while (!list_empty(&bdi->work_list));
1031 } else {
1032 /*
1033 * bdi_wq can't get enough workers and we're running off
1034 * the emergency worker. Don't hog it. Hopefully, 1024 is
1035 * enough for efficient IO.
1036 */
1037 pages_written = writeback_inodes_wb(&bdi->wb, 1024,
1038 WB_REASON_FORKER_THREAD);
Dave Chinner455b2862010-07-07 13:24:06 +10001039 trace_writeback_pages_written(pages_written);
Jens Axboe03ba3782009-09-09 09:08:54 +02001040 }
1041
Tejun Heo839a8e82013-04-01 19:08:06 -07001042 if (!list_empty(&bdi->work_list) ||
1043 (wb_has_dirty_io(wb) && dirty_writeback_interval))
1044 queue_delayed_work(bdi_wq, &wb->dwork,
1045 msecs_to_jiffies(dirty_writeback_interval * 10));
Dave Chinner455b2862010-07-07 13:24:06 +10001046
Tejun Heo839a8e82013-04-01 19:08:06 -07001047 current->flags &= ~PF_SWAPWRITE;
Jens Axboe03ba3782009-09-09 09:08:54 +02001048}
1049
1050/*
Jens Axboe03ba3782009-09-09 09:08:54 +02001051 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1052 * the whole world.
1053 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001054void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
Jens Axboe03ba3782009-09-09 09:08:54 +02001055{
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001056 struct backing_dev_info *bdi;
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001057
Jan Kara47df3dd2013-09-11 14:22:22 -07001058 if (!nr_pages)
1059 nr_pages = get_nr_dirty_pages();
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001060
1061 rcu_read_lock();
1062 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1063 if (!bdi_has_dirty_io(bdi))
1064 continue;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001065 __bdi_start_writeback(bdi, nr_pages, false, reason);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001066 }
1067 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +02001068}
1069
1070static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1071{
1072 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1073 struct dentry *dentry;
1074 const char *name = "?";
1075
1076 dentry = d_find_alias(inode);
1077 if (dentry) {
1078 spin_lock(&dentry->d_lock);
1079 name = (const char *) dentry->d_name.name;
1080 }
1081 printk(KERN_DEBUG
1082 "%s(%d): dirtied inode %lu (%s) on %s\n",
1083 current->comm, task_pid_nr(current), inode->i_ino,
1084 name, inode->i_sb->s_id);
1085 if (dentry) {
1086 spin_unlock(&dentry->d_lock);
1087 dput(dentry);
1088 }
1089 }
1090}
1091
1092/**
1093 * __mark_inode_dirty - internal function
1094 * @inode: inode to mark
1095 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1096 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1097 * mark_inode_dirty_sync.
1098 *
1099 * Put the inode on the super block's dirty list.
1100 *
1101 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1102 * dirty list only if it is hashed or if it refers to a blockdev.
1103 * If it was not hashed, it will never be added to the dirty list
1104 * even if it is later hashed, as it will have been marked dirty already.
1105 *
1106 * In short, make sure you hash any inodes _before_ you start marking
1107 * them dirty.
1108 *
Jens Axboe03ba3782009-09-09 09:08:54 +02001109 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1110 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1111 * the kernel-internal blockdev inode represents the dirtying time of the
1112 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1113 * page->mapping->host, so the page-dirtying time is recorded in the internal
1114 * blockdev inode.
1115 */
1116void __mark_inode_dirty(struct inode *inode, int flags)
1117{
1118 struct super_block *sb = inode->i_sb;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001119 struct backing_dev_info *bdi = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +02001120
1121 /*
1122 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1123 * dirty the inode itself
1124 */
1125 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Tejun Heo9fb0a7d2013-01-11 13:06:37 -08001126 trace_writeback_dirty_inode_start(inode, flags);
1127
Jens Axboe03ba3782009-09-09 09:08:54 +02001128 if (sb->s_op->dirty_inode)
Christoph Hellwigaa385722011-05-27 06:53:02 -04001129 sb->s_op->dirty_inode(inode, flags);
Tejun Heo9fb0a7d2013-01-11 13:06:37 -08001130
1131 trace_writeback_dirty_inode(inode, flags);
Jens Axboe03ba3782009-09-09 09:08:54 +02001132 }
1133
1134 /*
1135 * make sure that changes are seen by all cpus before we test i_state
1136 * -- mikulas
1137 */
1138 smp_mb();
1139
1140 /* avoid the locking if we can */
1141 if ((inode->i_state & flags) == flags)
1142 return;
1143
1144 if (unlikely(block_dump))
1145 block_dump___mark_inode_dirty(inode);
1146
Dave Chinner250df6e2011-03-22 22:23:36 +11001147 spin_lock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001148 if ((inode->i_state & flags) != flags) {
1149 const int was_dirty = inode->i_state & I_DIRTY;
1150
1151 inode->i_state |= flags;
1152
1153 /*
1154 * If the inode is being synced, just update its dirty state.
1155 * The unlocker will place the inode on the appropriate
1156 * superblock list, based upon its state.
1157 */
1158 if (inode->i_state & I_SYNC)
Dave Chinner250df6e2011-03-22 22:23:36 +11001159 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001160
1161 /*
1162 * Only add valid (hashed) inodes to the superblock's
1163 * dirty list. Add blockdev inodes as well.
1164 */
1165 if (!S_ISBLK(inode->i_mode)) {
Al Viro1d3382cb2010-10-23 15:19:20 -04001166 if (inode_unhashed(inode))
Dave Chinner250df6e2011-03-22 22:23:36 +11001167 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001168 }
Al Viroa4ffdde2010-06-02 17:38:30 -04001169 if (inode->i_state & I_FREEING)
Dave Chinner250df6e2011-03-22 22:23:36 +11001170 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001171
1172 /*
1173 * If the inode was already on b_dirty/b_io/b_more_io, don't
1174 * reposition it (that would break b_dirty time-ordering).
1175 */
1176 if (!was_dirty) {
Dave Chinnera66979a2011-03-22 22:23:41 +11001177 bool wakeup_bdi = false;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001178 bdi = inode_to_bdi(inode);
Jens Axboe500b0672009-09-09 09:10:25 +02001179
Junxiao Bi146d7002013-09-11 14:23:04 -07001180 spin_unlock(&inode->i_lock);
1181 spin_lock(&bdi->wb.list_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001182 if (bdi_cap_writeback_dirty(bdi)) {
1183 WARN(!test_bit(BDI_registered, &bdi->state),
1184 "bdi-%s not registered\n", bdi->name);
1185
1186 /*
1187 * If this is the first dirty inode for this
1188 * bdi, we have to wake-up the corresponding
1189 * bdi thread to make sure background
1190 * write-back happens later.
1191 */
1192 if (!wb_has_dirty_io(&bdi->wb))
1193 wakeup_bdi = true;
Jens Axboe500b0672009-09-09 09:10:25 +02001194 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001195
1196 inode->dirtied_when = jiffies;
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001197 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001198 spin_unlock(&bdi->wb.list_lock);
Dave Chinnera66979a2011-03-22 22:23:41 +11001199
1200 if (wakeup_bdi)
1201 bdi_wakeup_thread_delayed(bdi);
1202 return;
Jens Axboe03ba3782009-09-09 09:08:54 +02001203 }
1204 }
Dave Chinner250df6e2011-03-22 22:23:36 +11001205out_unlock_inode:
1206 spin_unlock(&inode->i_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001207
Jens Axboe03ba3782009-09-09 09:08:54 +02001208}
1209EXPORT_SYMBOL(__mark_inode_dirty);
1210
Jens Axboeb6e51312009-09-16 15:13:54 +02001211static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001212{
Jens Axboe03ba3782009-09-09 09:08:54 +02001213 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001214
Jens Axboe03ba3782009-09-09 09:08:54 +02001215 /*
1216 * We need to be protected against the filesystem going from
1217 * r/o to r/w or vice versa.
1218 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001219 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001220
Dave Chinner55fa6092011-03-22 22:23:40 +11001221 spin_lock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001222
1223 /*
1224 * Data integrity sync. Must wait for all pages under writeback,
1225 * because there may have been pages dirtied before our sync
1226 * call, but which had writeout started before we write it out.
1227 * In which case, the inode may not be on the dirty list, but
1228 * we still have to wait for that writeout.
1229 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001230 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001231 struct address_space *mapping = inode->i_mapping;
Jens Axboe03ba3782009-09-09 09:08:54 +02001232
Dave Chinner250df6e2011-03-22 22:23:36 +11001233 spin_lock(&inode->i_lock);
1234 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1235 (mapping->nrpages == 0)) {
1236 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001237 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001238 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001239 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001240 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001241 spin_unlock(&inode_sb_list_lock);
1242
Jens Axboe03ba3782009-09-09 09:08:54 +02001243 /*
Dave Chinner55fa6092011-03-22 22:23:40 +11001244 * We hold a reference to 'inode' so it couldn't have been
1245 * removed from s_inodes list while we dropped the
1246 * inode_sb_list_lock. We cannot iput the inode now as we can
1247 * be holding the last reference and we cannot iput it under
1248 * inode_sb_list_lock. So we keep the reference and iput it
1249 * later.
Jens Axboe03ba3782009-09-09 09:08:54 +02001250 */
1251 iput(old_inode);
1252 old_inode = inode;
1253
1254 filemap_fdatawait(mapping);
1255
1256 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001257
Dave Chinner55fa6092011-03-22 22:23:40 +11001258 spin_lock(&inode_sb_list_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001259 }
Dave Chinner55fa6092011-03-22 22:23:40 +11001260 spin_unlock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001261 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262}
1263
Jens Axboed8a85592009-09-02 12:34:32 +02001264/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001265 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
Jens Axboed8a85592009-09-02 12:34:32 +02001266 * @sb: the superblock
Chris Mason3259f8b2010-10-29 11:16:17 -04001267 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001268 * @reason: reason why some writeback work initiated
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 *
Jens Axboed8a85592009-09-02 12:34:32 +02001270 * Start writeback on some inodes on this super_block. No guarantees are made
1271 * on how many (if any) will be written, and this function does not wait
Chris Mason3259f8b2010-10-29 11:16:17 -04001272 * for IO completion of submitted IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001274void writeback_inodes_sb_nr(struct super_block *sb,
1275 unsigned long nr,
1276 enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001278 DECLARE_COMPLETION_ONSTACK(done);
1279 struct wb_writeback_work work = {
Wu Fengguang6e6938b2010-06-06 10:38:15 -06001280 .sb = sb,
1281 .sync_mode = WB_SYNC_NONE,
1282 .tagged_writepages = 1,
1283 .done = &done,
1284 .nr_pages = nr,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001285 .reason = reason,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001286 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001287
Jan Kara6eedc702012-07-03 16:45:27 +02001288 if (sb->s_bdi == &noop_backing_dev_info)
1289 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001290 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001291 bdi_queue_work(sb->s_bdi, &work);
1292 wait_for_completion(&done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293}
Chris Mason3259f8b2010-10-29 11:16:17 -04001294EXPORT_SYMBOL(writeback_inodes_sb_nr);
1295
1296/**
1297 * writeback_inodes_sb - writeback dirty inodes from given super_block
1298 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001299 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001300 *
1301 * Start writeback on some inodes on this super_block. No guarantees are made
1302 * on how many (if any) will be written, and this function does not wait
1303 * for IO completion of submitted IO.
1304 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001305void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001306{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001307 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001308}
Jens Axboed8a85592009-09-02 12:34:32 +02001309EXPORT_SYMBOL(writeback_inodes_sb);
1310
1311/**
Miao Xie10ee27a2013-01-10 13:47:57 +08001312 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
Chris Mason3259f8b2010-10-29 11:16:17 -04001313 * @sb: the superblock
1314 * @nr: the number of pages to write
Miao Xie10ee27a2013-01-10 13:47:57 +08001315 * @reason: the reason of writeback
Chris Mason3259f8b2010-10-29 11:16:17 -04001316 *
Miao Xie10ee27a2013-01-10 13:47:57 +08001317 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
Chris Mason3259f8b2010-10-29 11:16:17 -04001318 * Returns 1 if writeback was started, 0 if not.
1319 */
Miao Xie10ee27a2013-01-10 13:47:57 +08001320int try_to_writeback_inodes_sb_nr(struct super_block *sb,
1321 unsigned long nr,
1322 enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001323{
Miao Xie10ee27a2013-01-10 13:47:57 +08001324 if (writeback_in_progress(sb->s_bdi))
Chris Mason3259f8b2010-10-29 11:16:17 -04001325 return 1;
Miao Xie10ee27a2013-01-10 13:47:57 +08001326
1327 if (!down_read_trylock(&sb->s_umount))
Chris Mason3259f8b2010-10-29 11:16:17 -04001328 return 0;
Miao Xie10ee27a2013-01-10 13:47:57 +08001329
1330 writeback_inodes_sb_nr(sb, nr, reason);
1331 up_read(&sb->s_umount);
1332 return 1;
Chris Mason3259f8b2010-10-29 11:16:17 -04001333}
Miao Xie10ee27a2013-01-10 13:47:57 +08001334EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr);
1335
1336/**
1337 * try_to_writeback_inodes_sb - try to start writeback if none underway
1338 * @sb: the superblock
1339 * @reason: reason why some writeback work was initiated
1340 *
1341 * Implement by try_to_writeback_inodes_sb_nr()
1342 * Returns 1 if writeback was started, 0 if not.
1343 */
1344int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1345{
1346 return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1347}
1348EXPORT_SYMBOL(try_to_writeback_inodes_sb);
Chris Mason3259f8b2010-10-29 11:16:17 -04001349
1350/**
Jens Axboed8a85592009-09-02 12:34:32 +02001351 * sync_inodes_sb - sync sb inode pages
Jan Kara0dc83bd2014-02-21 11:19:04 +01001352 * @sb: the superblock
Jens Axboed8a85592009-09-02 12:34:32 +02001353 *
1354 * This function writes and waits on any dirty inode belonging to this
Jan Kara0dc83bd2014-02-21 11:19:04 +01001355 * super_block.
Jens Axboed8a85592009-09-02 12:34:32 +02001356 */
Jan Kara0dc83bd2014-02-21 11:19:04 +01001357void sync_inodes_sb(struct super_block *sb)
Jens Axboed8a85592009-09-02 12:34:32 +02001358{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001359 DECLARE_COMPLETION_ONSTACK(done);
1360 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001361 .sb = sb,
1362 .sync_mode = WB_SYNC_ALL,
1363 .nr_pages = LONG_MAX,
1364 .range_cyclic = 0,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001365 .done = &done,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001366 .reason = WB_REASON_SYNC,
Dave Chinner7747bd42013-07-02 22:38:35 +10001367 .for_sync = 1,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001368 };
1369
Jan Kara6eedc702012-07-03 16:45:27 +02001370 /* Nothing to do? */
1371 if (sb->s_bdi == &noop_backing_dev_info)
1372 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001373 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1374
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001375 bdi_queue_work(sb->s_bdi, &work);
1376 wait_for_completion(&done);
1377
Jens Axboeb6e51312009-09-16 15:13:54 +02001378 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001379}
1380EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001383 * write_inode_now - write an inode to disk
1384 * @inode: inode to write to disk
1385 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001387 * This function commits an inode to disk immediately if it is dirty. This is
1388 * primarily needed by knfsd.
1389 *
1390 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392int write_inode_now(struct inode *inode, int sync)
1393{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001394 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 struct writeback_control wbc = {
1396 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001397 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001398 .range_start = 0,
1399 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 };
1401
1402 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001403 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404
1405 might_sleep();
Jan Kara4f8ad652012-05-03 14:48:00 +02001406 return writeback_single_inode(inode, wb, &wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407}
1408EXPORT_SYMBOL(write_inode_now);
1409
1410/**
1411 * sync_inode - write an inode and its pages to disk.
1412 * @inode: the inode to sync
1413 * @wbc: controls the writeback mode
1414 *
1415 * sync_inode() will write an inode and its pages to disk. It will also
1416 * correctly update the inode on its superblock's dirty inode lists and will
1417 * update inode->i_state.
1418 *
1419 * The caller must have a ref on the inode.
1420 */
1421int sync_inode(struct inode *inode, struct writeback_control *wbc)
1422{
Jan Kara4f8ad652012-05-03 14:48:00 +02001423 return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424}
1425EXPORT_SYMBOL(sync_inode);
Christoph Hellwigc37650162010-10-06 10:48:20 +02001426
1427/**
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001428 * sync_inode_metadata - write an inode to disk
Christoph Hellwigc37650162010-10-06 10:48:20 +02001429 * @inode: the inode to sync
1430 * @wait: wait for I/O to complete.
1431 *
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001432 * Write an inode to disk and adjust its dirty state after completion.
Christoph Hellwigc37650162010-10-06 10:48:20 +02001433 *
1434 * Note: only writes the actual inode, no associated data or other metadata.
1435 */
1436int sync_inode_metadata(struct inode *inode, int wait)
1437{
1438 struct writeback_control wbc = {
1439 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1440 .nr_to_write = 0, /* metadata-only */
1441 };
1442
1443 return sync_inode(inode, &wbc);
1444}
1445EXPORT_SYMBOL(sync_inode_metadata);