blob: 1f4a10ece2f1249d5ae9991bfdde4874019b3462 [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 Karac4a391b2013-11-12 15:07:51 -080043 /*
44 * Write only inodes dirtied before this time. Don't forget to set
45 * older_than_this_is_set when you set this.
46 */
47 unsigned long older_than_this;
Jens Axboec4a77a62009-09-16 15:18:25 +020048 enum writeback_sync_modes sync_mode;
Wu Fengguang6e6938b2010-06-06 10:38:15 -060049 unsigned int tagged_writepages:1;
H Hartley Sweeten52957fe2010-04-01 20:36:30 -050050 unsigned int for_kupdate:1;
51 unsigned int range_cyclic:1;
52 unsigned int for_background:1;
Dave Chinner7747bd42013-07-02 22:38:35 +100053 unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */
Jan Karac4a391b2013-11-12 15:07:51 -080054 unsigned int older_than_this_is_set:1;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -060055 enum wb_reason reason; /* why was writeback initiated? */
Jens Axboec4a77a62009-09-16 15:18:25 +020056
Jens Axboe8010c3b2009-09-15 20:04:57 +020057 struct list_head list; /* pending work list */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020058 struct completion *done; /* set if the caller waits */
Jens Axboe03ba3782009-09-09 09:08:54 +020059};
60
Adrian Bunkf11b00f2008-04-29 00:58:56 -070061/**
62 * writeback_in_progress - determine whether there is writeback in progress
63 * @bdi: the device's backing_dev_info structure.
64 *
Jens Axboe03ba3782009-09-09 09:08:54 +020065 * Determine whether there is writeback waiting to be handled against a
66 * backing device.
Adrian Bunkf11b00f2008-04-29 00:58:56 -070067 */
68int writeback_in_progress(struct backing_dev_info *bdi)
69{
Jan Kara81d73a32010-08-11 14:17:44 -070070 return test_bit(BDI_writeback_running, &bdi->state);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070071}
Theodore Ts'o00d4e732012-09-19 22:42:36 -040072EXPORT_SYMBOL(writeback_in_progress);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070073
Jan Kara692ebd12010-09-21 11:51:01 +020074static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
75{
76 struct super_block *sb = inode->i_sb;
Jan Kara692ebd12010-09-21 11:51:01 +020077
Jan Karaa8855992013-07-09 22:36:45 +080078 if (sb_is_blkdev_sb(sb))
Christoph Hellwigaaead252010-10-04 14:25:33 +020079 return inode->i_mapping->backing_dev_info;
80
81 return sb->s_bdi;
Jan Kara692ebd12010-09-21 11:51:01 +020082}
83
Nick Piggin7ccf19a2010-10-21 11:49:30 +110084static inline struct inode *wb_inode(struct list_head *head)
85{
86 return list_entry(head, struct inode, i_wb_list);
87}
88
Wu Fengguang15eb77a2012-01-17 11:18:56 -060089/*
90 * Include the creation of the trace points after defining the
91 * wb_writeback_work structure and inline functions so that the definition
92 * remains local to this file.
93 */
94#define CREATE_TRACE_POINTS
95#include <trace/events/writeback.h>
96
Jan Kara65850272011-01-13 15:45:44 -080097static void bdi_queue_work(struct backing_dev_info *bdi,
98 struct wb_writeback_work *work)
99{
100 trace_writeback_queue(bdi, work);
101
102 spin_lock_bh(&bdi->wb_lock);
103 list_add_tail(&work->list, &bdi->work_list);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300104 spin_unlock_bh(&bdi->wb_lock);
Tejun Heo839a8e82013-04-01 19:08:06 -0700105
106 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107}
108
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200109static void
110__bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600111 bool range_cyclic, enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200113 struct wb_writeback_work *work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200114
Jens Axboebcddc3f2009-09-13 20:07:36 +0200115 /*
116 * This is WB_SYNC_NONE writeback, so if allocation fails just
117 * wakeup the thread for old dirty data writeback
118 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200119 work = kzalloc(sizeof(*work), GFP_ATOMIC);
120 if (!work) {
Tejun Heo839a8e82013-04-01 19:08:06 -0700121 trace_writeback_nowork(bdi);
122 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200123 return;
Jens Axboebcddc3f2009-09-13 20:07:36 +0200124 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200125
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200126 work->sync_mode = WB_SYNC_NONE;
127 work->nr_pages = nr_pages;
128 work->range_cyclic = range_cyclic;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600129 work->reason = reason;
Christoph Hellwigf0fad8a2009-09-11 09:47:56 +0200130
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200131 bdi_queue_work(bdi, work);
Jens Axboeb6e51312009-09-16 15:13:54 +0200132}
133
134/**
135 * bdi_start_writeback - start writeback
136 * @bdi: the backing device to write from
137 * @nr_pages: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +0800138 * @reason: reason why some writeback work was initiated
Jens Axboeb6e51312009-09-16 15:13:54 +0200139 *
140 * Description:
141 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300142 * started when this function returns, we make no guarantees on
Jens Axboe0e3c9a22010-06-01 11:08:43 +0200143 * completion. Caller need not hold sb s_umount semaphore.
Jens Axboeb6e51312009-09-16 15:13:54 +0200144 *
145 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600146void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
147 enum wb_reason reason)
Jens Axboeb6e51312009-09-16 15:13:54 +0200148{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600149 __bdi_start_writeback(bdi, nr_pages, true, reason);
Christoph Hellwigc5444192010-06-08 18:15:15 +0200150}
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800151
Christoph Hellwigc5444192010-06-08 18:15:15 +0200152/**
153 * bdi_start_background_writeback - start background writeback
154 * @bdi: the backing device to write from
155 *
156 * Description:
Jan Kara65850272011-01-13 15:45:44 -0800157 * This makes sure WB_SYNC_NONE background writeback happens. When
158 * this function returns, it is only guaranteed that for given BDI
159 * some IO is happening if we are over background dirty threshold.
160 * Caller need not hold sb s_umount semaphore.
Christoph Hellwigc5444192010-06-08 18:15:15 +0200161 */
162void bdi_start_background_writeback(struct backing_dev_info *bdi)
163{
Jan Kara65850272011-01-13 15:45:44 -0800164 /*
165 * We just wake up the flusher thread. It will perform background
166 * writeback as soon as there is no other work to do.
167 */
Wu Fengguang71927e82011-01-13 15:45:46 -0800168 trace_writeback_wake_background(bdi);
Tejun Heo839a8e82013-04-01 19:08:06 -0700169 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170}
171
172/*
Dave Chinnera66979a2011-03-22 22:23:41 +1100173 * Remove the inode from the writeback list it is on.
174 */
175void inode_wb_list_del(struct inode *inode)
176{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600177 struct backing_dev_info *bdi = inode_to_bdi(inode);
Dave Chinnera66979a2011-03-22 22:23:41 +1100178
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600179 spin_lock(&bdi->wb.list_lock);
180 list_del_init(&inode->i_wb_list);
181 spin_unlock(&bdi->wb.list_lock);
182}
Dave Chinnera66979a2011-03-22 22:23:41 +1100183
184/*
Andrew Morton6610a0b2007-10-16 23:30:32 -0700185 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
186 * furthest end of its superblock's dirty-inode list.
187 *
188 * Before stamping the inode's ->dirtied_when, we check to see whether it is
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200189 * already the most-recently-dirtied inode on the b_dirty list. If that is
Andrew Morton6610a0b2007-10-16 23:30:32 -0700190 * the case then the inode must have been redirtied while it was being written
191 * out and we don't reset its dirtied_when.
192 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600193static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
Andrew Morton6610a0b2007-10-16 23:30:32 -0700194{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600195 assert_spin_locked(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200196 if (!list_empty(&wb->b_dirty)) {
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200197 struct inode *tail;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700198
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100199 tail = wb_inode(wb->b_dirty.next);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200200 if (time_before(inode->dirtied_when, tail->dirtied_when))
Andrew Morton6610a0b2007-10-16 23:30:32 -0700201 inode->dirtied_when = jiffies;
202 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100203 list_move(&inode->i_wb_list, &wb->b_dirty);
Andrew Morton6610a0b2007-10-16 23:30:32 -0700204}
205
206/*
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200207 * requeue inode for re-scanning after bdi->b_io list is exhausted.
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700208 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600209static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700210{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600211 assert_spin_locked(&wb->list_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100212 list_move(&inode->i_wb_list, &wb->b_more_io);
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700213}
214
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700215static void inode_sync_complete(struct inode *inode)
216{
Jan Kara365b94ae2012-05-03 14:47:55 +0200217 inode->i_state &= ~I_SYNC;
Jan Kara4eff96d2012-11-26 16:29:51 -0800218 /* If inode is clean an unused, put it into LRU now... */
219 inode_add_lru(inode);
Jan Kara365b94ae2012-05-03 14:47:55 +0200220 /* Waiters must see I_SYNC cleared before being woken up */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700221 smp_mb();
222 wake_up_bit(&inode->i_state, __I_SYNC);
223}
224
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700225static bool inode_dirtied_after(struct inode *inode, unsigned long t)
226{
227 bool ret = time_after(inode->dirtied_when, t);
228#ifndef CONFIG_64BIT
229 /*
230 * For inodes being constantly redirtied, dirtied_when can get stuck.
231 * It _appears_ to be in the future, but is actually in distant past.
232 * This test is necessary to prevent such wrapped-around relative times
Jens Axboe5b0830c2009-09-23 19:37:09 +0200233 * from permanently stopping the whole bdi writeback.
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700234 */
235 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
236#endif
237 return ret;
238}
239
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700240/*
Wang Sheng-Hui0e2f2b22012-09-11 08:28:18 +0800241 * Move expired (dirtied before work->older_than_this) dirty inodes from
Jan Kara697e6fe2012-03-09 07:26:22 -0800242 * @delaying_queue to @dispatch_queue.
Fengguang Wu2c136572007-10-16 23:30:39 -0700243 */
Wu Fengguange84d0a42011-04-23 12:27:27 -0600244static int move_expired_inodes(struct list_head *delaying_queue,
Fengguang Wu2c136572007-10-16 23:30:39 -0700245 struct list_head *dispatch_queue,
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600246 struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700247{
Shaohua Li5c034492009-09-24 14:42:33 +0200248 LIST_HEAD(tmp);
249 struct list_head *pos, *node;
Jens Axboecf137302009-09-24 15:12:57 +0200250 struct super_block *sb = NULL;
Shaohua Li5c034492009-09-24 14:42:33 +0200251 struct inode *inode;
Jens Axboecf137302009-09-24 15:12:57 +0200252 int do_sb_sort = 0;
Wu Fengguange84d0a42011-04-23 12:27:27 -0600253 int moved = 0;
Shaohua Li5c034492009-09-24 14:42:33 +0200254
Jan Karac4a391b2013-11-12 15:07:51 -0800255 WARN_ON_ONCE(!work->older_than_this_is_set);
Fengguang Wu2c136572007-10-16 23:30:39 -0700256 while (!list_empty(delaying_queue)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100257 inode = wb_inode(delaying_queue->prev);
Jan Karac4a391b2013-11-12 15:07:51 -0800258 if (inode_dirtied_after(inode, work->older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700259 break;
Jan Karaa8855992013-07-09 22:36:45 +0800260 list_move(&inode->i_wb_list, &tmp);
261 moved++;
262 if (sb_is_blkdev_sb(inode->i_sb))
263 continue;
Jens Axboecf137302009-09-24 15:12:57 +0200264 if (sb && sb != inode->i_sb)
265 do_sb_sort = 1;
266 sb = inode->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200267 }
268
Jens Axboecf137302009-09-24 15:12:57 +0200269 /* just one sb in list, splice to dispatch_queue and we're done */
270 if (!do_sb_sort) {
271 list_splice(&tmp, dispatch_queue);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600272 goto out;
Jens Axboecf137302009-09-24 15:12:57 +0200273 }
274
Shaohua Li5c034492009-09-24 14:42:33 +0200275 /* Move inodes from one superblock together */
276 while (!list_empty(&tmp)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100277 sb = wb_inode(tmp.prev)->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200278 list_for_each_prev_safe(pos, node, &tmp) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100279 inode = wb_inode(pos);
Shaohua Li5c034492009-09-24 14:42:33 +0200280 if (inode->i_sb == sb)
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100281 list_move(&inode->i_wb_list, dispatch_queue);
Shaohua Li5c034492009-09-24 14:42:33 +0200282 }
Fengguang Wu2c136572007-10-16 23:30:39 -0700283 }
Wu Fengguange84d0a42011-04-23 12:27:27 -0600284out:
285 return moved;
Fengguang Wu2c136572007-10-16 23:30:39 -0700286}
287
288/*
289 * Queue all expired dirty inodes for io, eldest first.
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700290 * Before
291 * newly dirtied b_dirty b_io b_more_io
292 * =============> gf edc BA
293 * After
294 * newly dirtied b_dirty b_io b_more_io
295 * =============> g fBAedc
296 * |
297 * +--> dequeue for IO
Fengguang Wu2c136572007-10-16 23:30:39 -0700298 */
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600299static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700300{
Wu Fengguange84d0a42011-04-23 12:27:27 -0600301 int moved;
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600302 assert_spin_locked(&wb->list_lock);
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700303 list_splice_init(&wb->b_more_io, &wb->b_io);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600304 moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
305 trace_writeback_queue_io(wb, work, moved);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200306}
307
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100308static int write_inode(struct inode *inode, struct writeback_control *wbc)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200309{
Tejun Heo9fb0a7d2013-01-11 13:06:37 -0800310 int ret;
311
312 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) {
313 trace_writeback_write_inode_start(inode, wbc);
314 ret = inode->i_sb->s_op->write_inode(inode, wbc);
315 trace_writeback_write_inode(inode, wbc);
316 return ret;
317 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200318 return 0;
Fengguang Wu2c136572007-10-16 23:30:39 -0700319}
320
321/*
Jan Kara169ebd92012-05-03 14:48:03 +0200322 * Wait for writeback on an inode to complete. Called with i_lock held.
323 * Caller must make sure inode cannot go away when we drop i_lock.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200324 */
Jan Kara169ebd92012-05-03 14:48:03 +0200325static void __inode_wait_for_writeback(struct inode *inode)
326 __releases(inode->i_lock)
327 __acquires(inode->i_lock)
Christoph Hellwig01c03192009-06-08 13:35:40 +0200328{
329 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
330 wait_queue_head_t *wqh;
331
332 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100333 while (inode->i_state & I_SYNC) {
334 spin_unlock(&inode->i_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200335 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
Dave Chinner250df6e2011-03-22 22:23:36 +1100336 spin_lock(&inode->i_lock);
Richard Kennedy58a9d3d82010-05-24 14:32:38 -0700337 }
Christoph Hellwig01c03192009-06-08 13:35:40 +0200338}
339
340/*
Jan Kara169ebd92012-05-03 14:48:03 +0200341 * Wait for writeback on an inode to complete. Caller must have inode pinned.
342 */
343void inode_wait_for_writeback(struct inode *inode)
344{
345 spin_lock(&inode->i_lock);
346 __inode_wait_for_writeback(inode);
347 spin_unlock(&inode->i_lock);
348}
349
350/*
351 * Sleep until I_SYNC is cleared. This function must be called with i_lock
352 * held and drops it. It is aimed for callers not holding any inode reference
353 * so once i_lock is dropped, inode can go away.
354 */
355static void inode_sleep_on_writeback(struct inode *inode)
356 __releases(inode->i_lock)
357{
358 DEFINE_WAIT(wait);
359 wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
360 int sleep;
361
362 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
363 sleep = inode->i_state & I_SYNC;
364 spin_unlock(&inode->i_lock);
365 if (sleep)
366 schedule();
367 finish_wait(wqh, &wait);
368}
369
370/*
Jan Karaccb26b52012-05-03 14:47:58 +0200371 * Find proper writeback list for the inode depending on its current state and
372 * possibly also change of its state while we were doing writeback. Here we
373 * handle things such as livelock prevention or fairness of writeback among
374 * inodes. This function can be called only by flusher thread - noone else
375 * processes all inodes in writeback lists and requeueing inodes behind flusher
376 * thread's back can have unexpected consequences.
377 */
378static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
379 struct writeback_control *wbc)
380{
381 if (inode->i_state & I_FREEING)
382 return;
383
384 /*
385 * Sync livelock prevention. Each inode is tagged and synced in one
386 * shot. If still dirty, it will be redirty_tail()'ed below. Update
387 * the dirty time to prevent enqueue and sync it again.
388 */
389 if ((inode->i_state & I_DIRTY) &&
390 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
391 inode->dirtied_when = jiffies;
392
Jan Kara4f8ad652012-05-03 14:48:00 +0200393 if (wbc->pages_skipped) {
394 /*
395 * writeback is not making progress due to locked
396 * buffers. Skip this inode for now.
397 */
398 redirty_tail(inode, wb);
399 return;
400 }
401
Jan Karaccb26b52012-05-03 14:47:58 +0200402 if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
403 /*
404 * We didn't write back all the pages. nfs_writepages()
405 * sometimes bales out without doing anything.
406 */
407 if (wbc->nr_to_write <= 0) {
408 /* Slice used up. Queue for next turn. */
409 requeue_io(inode, wb);
410 } else {
411 /*
412 * Writeback blocked by something other than
413 * congestion. Delay the inode for some time to
414 * avoid spinning on the CPU (100% iowait)
415 * retrying writeback of the dirty page/inode
416 * that cannot be performed immediately.
417 */
418 redirty_tail(inode, wb);
419 }
420 } else if (inode->i_state & I_DIRTY) {
421 /*
422 * Filesystems can dirty the inode during writeback operations,
423 * such as delayed allocation during submission or metadata
424 * updates after data IO completion.
425 */
426 redirty_tail(inode, wb);
427 } else {
428 /* The inode is clean. Remove from writeback lists. */
429 list_del_init(&inode->i_wb_list);
430 }
431}
432
433/*
Jan Kara4f8ad652012-05-03 14:48:00 +0200434 * Write out an inode and its dirty pages. Do not update the writeback list
435 * linkage. That is left to the caller. The caller is also responsible for
436 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 */
438static int
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700439__writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 struct address_space *mapping = inode->i_mapping;
Wu Fengguang251d6a42010-12-01 17:33:37 -0600442 long nr_to_write = wbc->nr_to_write;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200443 unsigned dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 int ret;
445
Jan Kara4f8ad652012-05-03 14:48:00 +0200446 WARN_ON(!(inode->i_state & I_SYNC));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447
Tejun Heo9fb0a7d2013-01-11 13:06:37 -0800448 trace_writeback_single_inode_start(inode, wbc, nr_to_write);
449
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 ret = do_writepages(mapping, wbc);
451
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100452 /*
453 * Make sure to wait on the data before writing out the metadata.
454 * This is important for filesystems that modify metadata on data
Dave Chinner7747bd42013-07-02 22:38:35 +1000455 * I/O completion. We don't do it for sync(2) writeback because it has a
456 * separate, external IO completion path and ->sync_fs for guaranteeing
457 * inode metadata is written back correctly.
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100458 */
Dave Chinner7747bd42013-07-02 22:38:35 +1000459 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) {
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100460 int err = filemap_fdatawait(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 if (ret == 0)
462 ret = err;
463 }
464
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400465 /*
466 * Some filesystems may redirty the inode during the writeback
467 * due to delalloc, clear dirty metadata flags right before
468 * write_inode()
469 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100470 spin_lock(&inode->i_lock);
Jan Kara6290be12012-05-03 14:47:57 +0200471 /* Clear I_DIRTY_PAGES if we've written out all dirty pages */
472 if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
473 inode->i_state &= ~I_DIRTY_PAGES;
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400474 dirty = inode->i_state & I_DIRTY;
475 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100476 spin_unlock(&inode->i_lock);
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100477 /* Don't write the inode if only I_DIRTY_PAGES was set */
478 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100479 int err = write_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 if (ret == 0)
481 ret = err;
482 }
Jan Kara4f8ad652012-05-03 14:48:00 +0200483 trace_writeback_single_inode(inode, wbc, nr_to_write);
484 return ret;
485}
486
487/*
488 * Write out an inode's dirty pages. Either the caller has an active reference
489 * on the inode or the inode has I_WILL_FREE set.
490 *
491 * This function is designed to be called for writing back one inode which
492 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
493 * and does more profound writeback list handling in writeback_sb_inodes().
494 */
495static int
496writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
497 struct writeback_control *wbc)
498{
499 int ret = 0;
500
501 spin_lock(&inode->i_lock);
502 if (!atomic_read(&inode->i_count))
503 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
504 else
505 WARN_ON(inode->i_state & I_WILL_FREE);
506
507 if (inode->i_state & I_SYNC) {
508 if (wbc->sync_mode != WB_SYNC_ALL)
509 goto out;
510 /*
Jan Kara169ebd92012-05-03 14:48:03 +0200511 * It's a data-integrity sync. We must wait. Since callers hold
512 * inode reference or inode has I_WILL_FREE set, it cannot go
513 * away under us.
Jan Kara4f8ad652012-05-03 14:48:00 +0200514 */
Jan Kara169ebd92012-05-03 14:48:03 +0200515 __inode_wait_for_writeback(inode);
Jan Kara4f8ad652012-05-03 14:48:00 +0200516 }
517 WARN_ON(inode->i_state & I_SYNC);
518 /*
519 * Skip inode if it is clean. We don't want to mess with writeback
520 * lists in this function since flusher thread may be doing for example
521 * sync in parallel and if we move the inode, it could get skipped. So
522 * here we make sure inode is on some writeback list and leave it there
523 * unless we have completely cleaned the inode.
524 */
525 if (!(inode->i_state & I_DIRTY))
526 goto out;
527 inode->i_state |= I_SYNC;
528 spin_unlock(&inode->i_lock);
529
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700530 ret = __writeback_single_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600532 spin_lock(&wb->list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100533 spin_lock(&inode->i_lock);
Jan Kara4f8ad652012-05-03 14:48:00 +0200534 /*
535 * If inode is clean, remove it from writeback lists. Otherwise don't
536 * touch it. See comment above for explanation.
537 */
538 if (!(inode->i_state & I_DIRTY))
539 list_del_init(&inode->i_wb_list);
540 spin_unlock(&wb->list_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700541 inode_sync_complete(inode);
Jan Kara4f8ad652012-05-03 14:48:00 +0200542out:
543 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 return ret;
545}
546
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600547static long writeback_chunk_size(struct backing_dev_info *bdi,
548 struct wb_writeback_work *work)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600549{
550 long pages;
551
552 /*
553 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
554 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
555 * here avoids calling into writeback_inodes_wb() more than once.
556 *
557 * The intended call sequence for WB_SYNC_ALL writeback is:
558 *
559 * wb_writeback()
560 * writeback_sb_inodes() <== called only once
561 * write_cache_pages() <== called once for each inode
562 * (quickly) tag currently dirty pages
563 * (maybe slowly) sync all tagged pages
564 */
565 if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
566 pages = LONG_MAX;
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600567 else {
568 pages = min(bdi->avg_write_bandwidth / 2,
569 global_dirty_limit / DIRTY_SCOPE);
570 pages = min(pages, work->nr_pages);
571 pages = round_down(pages + MIN_WRITEBACK_PAGES,
572 MIN_WRITEBACK_PAGES);
573 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600574
575 return pages;
576}
577
Jens Axboe03ba3782009-09-09 09:08:54 +0200578/*
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800579 * Write a portion of b_io inodes which belong to @sb.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200580 *
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600581 * Return the number of pages and/or inodes written.
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800582 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600583static long writeback_sb_inodes(struct super_block *sb,
584 struct bdi_writeback *wb,
585 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200586{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600587 struct writeback_control wbc = {
588 .sync_mode = work->sync_mode,
589 .tagged_writepages = work->tagged_writepages,
590 .for_kupdate = work->for_kupdate,
591 .for_background = work->for_background,
Dave Chinner7747bd42013-07-02 22:38:35 +1000592 .for_sync = work->for_sync,
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600593 .range_cyclic = work->range_cyclic,
594 .range_start = 0,
595 .range_end = LLONG_MAX,
596 };
597 unsigned long start_time = jiffies;
598 long write_chunk;
599 long wrote = 0; /* count both pages and inodes */
600
Jens Axboe03ba3782009-09-09 09:08:54 +0200601 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100602 struct inode *inode = wb_inode(wb->b_io.prev);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200603
604 if (inode->i_sb != sb) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600605 if (work->sb) {
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200606 /*
607 * We only want to write back data for this
608 * superblock, move all inodes not belonging
609 * to it back onto the dirty list.
610 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600611 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200612 continue;
613 }
614
615 /*
616 * The inode belongs to a different superblock.
617 * Bounce back to the caller to unpin this and
618 * pin the next superblock.
619 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600620 break;
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200621 }
622
Christoph Hellwig9843b762010-10-24 19:40:46 +0200623 /*
Wanpeng Li331cbde2012-06-09 11:10:55 +0800624 * Don't bother with new inodes or inodes being freed, first
625 * kind does not need periodic writeout yet, and for the latter
Christoph Hellwig9843b762010-10-24 19:40:46 +0200626 * kind writeout is handled by the freer.
627 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100628 spin_lock(&inode->i_lock);
Christoph Hellwig9843b762010-10-24 19:40:46 +0200629 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100630 spin_unlock(&inode->i_lock);
Wu Fengguangfcc5c222011-07-11 23:08:50 -0700631 redirty_tail(inode, wb);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700632 continue;
633 }
Jan Karacc1676d2012-05-03 14:47:56 +0200634 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
635 /*
636 * If this inode is locked for writeback and we are not
637 * doing writeback-for-data-integrity, move it to
638 * b_more_io so that writeback can proceed with the
639 * other inodes on s_io.
640 *
641 * We'll have another go at writing back this inode
642 * when we completed a full scan of b_io.
643 */
644 spin_unlock(&inode->i_lock);
645 requeue_io(inode, wb);
646 trace_writeback_sb_inodes_requeue(inode);
647 continue;
648 }
Jan Karaf0d07b72012-05-03 14:47:59 +0200649 spin_unlock(&wb->list_lock);
650
Jan Kara4f8ad652012-05-03 14:48:00 +0200651 /*
652 * We already requeued the inode if it had I_SYNC set and we
653 * are doing WB_SYNC_NONE writeback. So this catches only the
654 * WB_SYNC_ALL case.
655 */
Jan Kara169ebd92012-05-03 14:48:03 +0200656 if (inode->i_state & I_SYNC) {
657 /* Wait for I_SYNC. This function drops i_lock... */
658 inode_sleep_on_writeback(inode);
659 /* Inode may be gone, start again */
Jan Karaead188f2012-06-08 17:07:36 +0200660 spin_lock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200661 continue;
662 }
Jan Kara4f8ad652012-05-03 14:48:00 +0200663 inode->i_state |= I_SYNC;
664 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200665
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600666 write_chunk = writeback_chunk_size(wb->bdi, work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600667 wbc.nr_to_write = write_chunk;
668 wbc.pages_skipped = 0;
Dave Chinner250df6e2011-03-22 22:23:36 +1100669
Jan Kara169ebd92012-05-03 14:48:03 +0200670 /*
671 * We use I_SYNC to pin the inode in memory. While it is set
672 * evict_inode() will wait so the inode cannot be freed.
673 */
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700674 __writeback_single_inode(inode, &wbc);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600675
676 work->nr_pages -= write_chunk - wbc.nr_to_write;
677 wrote += write_chunk - wbc.nr_to_write;
Jan Kara4f8ad652012-05-03 14:48:00 +0200678 spin_lock(&wb->list_lock);
679 spin_lock(&inode->i_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600680 if (!(inode->i_state & I_DIRTY))
681 wrote++;
Jan Kara4f8ad652012-05-03 14:48:00 +0200682 requeue_inode(inode, wb, &wbc);
683 inode_sync_complete(inode);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100684 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200685 cond_resched_lock(&wb->list_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600686 /*
687 * bail out to wb_writeback() often enough to check
688 * background threshold and other termination conditions.
689 */
690 if (wrote) {
691 if (time_is_before_jiffies(start_time + HZ / 10UL))
692 break;
693 if (work->nr_pages <= 0)
694 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800695 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600697 return wrote;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800698}
Nick Piggin38f21972009-01-06 14:40:25 -0800699
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600700static long __writeback_inodes_wb(struct bdi_writeback *wb,
701 struct wb_writeback_work *work)
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800702{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600703 unsigned long start_time = jiffies;
704 long wrote = 0;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800705
706 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100707 struct inode *inode = wb_inode(wb->b_io.prev);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800708 struct super_block *sb = inode->i_sb;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800709
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000710 if (!grab_super_passive(sb)) {
Wu Fengguang0e9958162011-07-29 22:14:35 -0600711 /*
712 * grab_super_passive() may fail consistently due to
713 * s_umount being grabbed by someone else. Don't use
714 * requeue_io() to avoid busy retrying the inode/sb.
715 */
716 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200717 continue;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800718 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600719 wrote += writeback_sb_inodes(sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200720 drop_super(sb);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800721
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600722 /* refer to the same tests at the end of writeback_sb_inodes */
723 if (wrote) {
724 if (time_is_before_jiffies(start_time + HZ / 10UL))
725 break;
726 if (work->nr_pages <= 0)
727 break;
728 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800729 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200730 /* Leave any unwritten inodes on b_io */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600731 return wrote;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200732}
733
Wanpeng Li7d9f0732013-09-11 14:22:40 -0700734static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600735 enum wb_reason reason)
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200736{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600737 struct wb_writeback_work work = {
738 .nr_pages = nr_pages,
739 .sync_mode = WB_SYNC_NONE,
740 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600741 .reason = reason,
Jan Karac4a391b2013-11-12 15:07:51 -0800742 .older_than_this = jiffies,
743 .older_than_this_is_set = 1,
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600744 };
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200745
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600746 spin_lock(&wb->list_lock);
Wu Fengguang424b3512010-07-21 20:11:53 -0600747 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600748 queue_io(wb, &work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600749 __writeback_inodes_wb(wb, &work);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600750 spin_unlock(&wb->list_lock);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200751
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600752 return nr_pages - work.nr_pages;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200753}
754
Wu Fengguangb00949a2010-11-18 14:38:33 -0600755static bool over_bground_thresh(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200756{
757 unsigned long background_thresh, dirty_thresh;
758
Wu Fengguang16c40422010-08-11 14:17:39 -0700759 global_dirty_limits(&background_thresh, &dirty_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200760
Wu Fengguangb00949a2010-11-18 14:38:33 -0600761 if (global_page_state(NR_FILE_DIRTY) +
762 global_page_state(NR_UNSTABLE_NFS) > background_thresh)
763 return true;
764
765 if (bdi_stat(bdi, BDI_RECLAIMABLE) >
766 bdi_dirty_limit(bdi, background_thresh))
767 return true;
768
769 return false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200770}
771
772/*
Wu Fengguange98be2d2010-08-29 11:22:30 -0600773 * Called under wb->list_lock. If there are multiple wb per bdi,
774 * only the flusher working on the first wb should do it.
775 */
776static void wb_update_bandwidth(struct bdi_writeback *wb,
777 unsigned long start_time)
778{
Wu Fengguangaf6a3112011-10-03 20:46:17 -0600779 __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
Wu Fengguange98be2d2010-08-29 11:22:30 -0600780}
781
782/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200783 * Explicit flushing or periodic writeback of "old" data.
784 *
785 * Define "old": the first time one of an inode's pages is dirtied, we mark the
786 * dirtying-time in the inode's address_space. So this periodic writeback code
787 * just walks the superblock inode list, writing back any inodes which are
788 * older than a specific point in time.
789 *
790 * Try to run once per dirty_writeback_interval. But if a writeback event
791 * takes longer than a dirty_writeback_interval interval, then leave a
792 * one-second gap.
793 *
794 * older_than_this takes precedence over nr_to_write. So we'll only write back
795 * all dirty pages if they are all attached to "old" mappings.
796 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200797static long wb_writeback(struct bdi_writeback *wb,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200798 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200799{
Wu Fengguange98be2d2010-08-29 11:22:30 -0600800 unsigned long wb_start = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600801 long nr_pages = work->nr_pages;
Jan Karaa5989bd2009-09-16 19:22:48 +0200802 struct inode *inode;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600803 long progress;
Jens Axboe03ba3782009-09-09 09:08:54 +0200804
Jan Karac4a391b2013-11-12 15:07:51 -0800805 if (!work->older_than_this_is_set) {
806 work->older_than_this = jiffies;
807 work->older_than_this_is_set = 1;
808 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200809
Wu Fengguange8dfc302011-04-21 12:06:32 -0600810 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200811 for (;;) {
812 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800813 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200814 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200815 if (work->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200816 break;
817
818 /*
Jan Karaaa373cf2011-01-13 15:45:47 -0800819 * Background writeout and kupdate-style writeback may
820 * run forever. Stop them if there is other work to do
821 * so that e.g. sync can proceed. They'll be restarted
822 * after the other works are all done.
823 */
824 if ((work->for_background || work->for_kupdate) &&
825 !list_empty(&wb->bdi->work_list))
826 break;
827
828 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800829 * For background writeout, stop when we are below the
830 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200831 */
Wu Fengguangb00949a2010-11-18 14:38:33 -0600832 if (work->for_background && !over_bground_thresh(wb->bdi))
Jens Axboe03ba3782009-09-09 09:08:54 +0200833 break;
834
Jan Kara1bc36b62011-10-19 11:44:41 +0200835 /*
836 * Kupdate and background works are special and we want to
837 * include all inodes that need writing. Livelock avoidance is
838 * handled by these works yielding to any other work so we are
839 * safe.
840 */
Wu Fengguangba9aa832010-07-21 20:32:30 -0600841 if (work->for_kupdate) {
Jan Karac4a391b2013-11-12 15:07:51 -0800842 work->older_than_this = jiffies -
Wu Fengguangba9aa832010-07-21 20:32:30 -0600843 msecs_to_jiffies(dirty_expire_interval * 10);
Jan Kara1bc36b62011-10-19 11:44:41 +0200844 } else if (work->for_background)
Jan Karac4a391b2013-11-12 15:07:51 -0800845 work->older_than_this = jiffies;
Dave Chinner028c2dd2010-07-07 13:24:07 +1000846
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600847 trace_writeback_start(wb->bdi, work);
Wu Fengguange8dfc302011-04-21 12:06:32 -0600848 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600849 queue_io(wb, work);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200850 if (work->sb)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600851 progress = writeback_sb_inodes(work->sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200852 else
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600853 progress = __writeback_inodes_wb(wb, work);
854 trace_writeback_written(wb->bdi, work);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000855
Wu Fengguange98be2d2010-08-29 11:22:30 -0600856 wb_update_bandwidth(wb, wb_start);
Jens Axboe03ba3782009-09-09 09:08:54 +0200857
858 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200859 * Did we write something? Try for more
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600860 *
861 * Dirty inodes are moved to b_io for writeback in batches.
862 * The completion of the current batch does not necessarily
863 * mean the overall work is done. So we keep looping as long
864 * as made some progress on cleaning pages or inodes.
Jens Axboe71fd05a2009-09-23 19:32:26 +0200865 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600866 if (progress)
Jens Axboe71fd05a2009-09-23 19:32:26 +0200867 continue;
868 /*
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600869 * No more inodes for IO, bail
Jens Axboe71fd05a2009-09-23 19:32:26 +0200870 */
Wu Fengguangb7a24412010-07-21 22:19:51 -0600871 if (list_empty(&wb->b_more_io))
Jens Axboe03ba3782009-09-09 09:08:54 +0200872 break;
873 /*
Jens Axboe8010c3b2009-09-15 20:04:57 +0200874 * Nothing written. Wait for some inode to
875 * become available for writeback. Otherwise
876 * we'll just busyloop.
Jens Axboe03ba3782009-09-09 09:08:54 +0200877 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200878 if (!list_empty(&wb->b_more_io)) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600879 trace_writeback_wait(wb->bdi, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200880 inode = wb_inode(wb->b_more_io.prev);
Dave Chinner250df6e2011-03-22 22:23:36 +1100881 spin_lock(&inode->i_lock);
Jan Karaf0d07b72012-05-03 14:47:59 +0200882 spin_unlock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200883 /* This function drops i_lock... */
884 inode_sleep_on_writeback(inode);
Jan Karaf0d07b72012-05-03 14:47:59 +0200885 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200886 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200887 }
Wu Fengguange8dfc302011-04-21 12:06:32 -0600888 spin_unlock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200889
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600890 return nr_pages - work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200891}
892
893/*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200894 * Return the next wb_writeback_work struct that hasn't been processed yet.
Jens Axboe03ba3782009-09-09 09:08:54 +0200895 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200896static struct wb_writeback_work *
Minchan Kim08852b62010-08-03 12:51:16 +0200897get_next_work_item(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200898{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200899 struct wb_writeback_work *work = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200900
Artem Bityutskiy64677162010-07-25 14:29:22 +0300901 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200902 if (!list_empty(&bdi->work_list)) {
903 work = list_entry(bdi->work_list.next,
904 struct wb_writeback_work, list);
905 list_del_init(&work->list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200906 }
Artem Bityutskiy64677162010-07-25 14:29:22 +0300907 spin_unlock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200908 return work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200909}
910
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700911/*
912 * Add in the number of potentially dirty inodes, because each inode
913 * write can dirty pagecache in the underlying blockdev.
914 */
915static unsigned long get_nr_dirty_pages(void)
916{
917 return global_page_state(NR_FILE_DIRTY) +
918 global_page_state(NR_UNSTABLE_NFS) +
919 get_nr_dirty_inodes();
920}
921
Jan Kara65850272011-01-13 15:45:44 -0800922static long wb_check_background_flush(struct bdi_writeback *wb)
923{
Wu Fengguangb00949a2010-11-18 14:38:33 -0600924 if (over_bground_thresh(wb->bdi)) {
Jan Kara65850272011-01-13 15:45:44 -0800925
926 struct wb_writeback_work work = {
927 .nr_pages = LONG_MAX,
928 .sync_mode = WB_SYNC_NONE,
929 .for_background = 1,
930 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600931 .reason = WB_REASON_BACKGROUND,
Jan Kara65850272011-01-13 15:45:44 -0800932 };
933
934 return wb_writeback(wb, &work);
935 }
936
937 return 0;
938}
939
Jens Axboe03ba3782009-09-09 09:08:54 +0200940static long wb_check_old_data_flush(struct bdi_writeback *wb)
941{
942 unsigned long expired;
943 long nr_pages;
944
Jens Axboe69b62d02010-05-17 12:51:03 +0200945 /*
946 * When set to zero, disable periodic writeback
947 */
948 if (!dirty_writeback_interval)
949 return 0;
950
Jens Axboe03ba3782009-09-09 09:08:54 +0200951 expired = wb->last_old_flush +
952 msecs_to_jiffies(dirty_writeback_interval * 10);
953 if (time_before(jiffies, expired))
954 return 0;
955
956 wb->last_old_flush = jiffies;
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700957 nr_pages = get_nr_dirty_pages();
Jens Axboe03ba3782009-09-09 09:08:54 +0200958
Jens Axboec4a77a62009-09-16 15:18:25 +0200959 if (nr_pages) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200960 struct wb_writeback_work work = {
Jens Axboec4a77a62009-09-16 15:18:25 +0200961 .nr_pages = nr_pages,
962 .sync_mode = WB_SYNC_NONE,
963 .for_kupdate = 1,
964 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600965 .reason = WB_REASON_PERIODIC,
Jens Axboec4a77a62009-09-16 15:18:25 +0200966 };
967
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200968 return wb_writeback(wb, &work);
Jens Axboec4a77a62009-09-16 15:18:25 +0200969 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200970
971 return 0;
972}
973
974/*
975 * Retrieve work items and do the writeback they describe
976 */
Wanpeng Li25d130b2013-07-08 16:00:14 -0700977static long wb_do_writeback(struct bdi_writeback *wb)
Jens Axboe03ba3782009-09-09 09:08:54 +0200978{
979 struct backing_dev_info *bdi = wb->bdi;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200980 struct wb_writeback_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +0200981 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200982
Jan Kara81d73a32010-08-11 14:17:44 -0700983 set_bit(BDI_writeback_running, &wb->bdi->state);
Minchan Kim08852b62010-08-03 12:51:16 +0200984 while ((work = get_next_work_item(bdi)) != NULL) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200985
Dave Chinner455b2862010-07-07 13:24:06 +1000986 trace_writeback_exec(bdi, work);
987
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200988 wrote += wb_writeback(wb, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200989
990 /*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200991 * Notify the caller of completion if this is a synchronous
992 * work item, otherwise just free it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200993 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200994 if (work->done)
995 complete(work->done);
996 else
997 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200998 }
999
1000 /*
1001 * Check for periodic writeback, kupdated() style
1002 */
1003 wrote += wb_check_old_data_flush(wb);
Jan Kara65850272011-01-13 15:45:44 -08001004 wrote += wb_check_background_flush(wb);
Jan Kara81d73a32010-08-11 14:17:44 -07001005 clear_bit(BDI_writeback_running, &wb->bdi->state);
Jens Axboe03ba3782009-09-09 09:08:54 +02001006
1007 return wrote;
1008}
1009
1010/*
1011 * Handle writeback of dirty data for the device backed by this bdi. Also
Tejun Heo839a8e82013-04-01 19:08:06 -07001012 * reschedules periodically and does kupdated style flushing.
Jens Axboe03ba3782009-09-09 09:08:54 +02001013 */
Tejun Heo839a8e82013-04-01 19:08:06 -07001014void bdi_writeback_workfn(struct work_struct *work)
Jens Axboe03ba3782009-09-09 09:08:54 +02001015{
Tejun Heo839a8e82013-04-01 19:08:06 -07001016 struct bdi_writeback *wb = container_of(to_delayed_work(work),
1017 struct bdi_writeback, dwork);
Christoph Hellwig08243902010-06-19 23:08:22 +02001018 struct backing_dev_info *bdi = wb->bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +02001019 long pages_written;
1020
Tejun Heoef3b1012013-04-30 15:27:24 -07001021 set_worker_desc("flush-%s", dev_name(bdi->dev));
Peter Zijlstra766f9162010-10-26 14:22:45 -07001022 current->flags |= PF_SWAPWRITE;
Christoph Hellwig08243902010-06-19 23:08:22 +02001023
Tejun Heo839a8e82013-04-01 19:08:06 -07001024 if (likely(!current_is_workqueue_rescuer() ||
1025 list_empty(&bdi->bdi_list))) {
Artem Bityutskiy64677162010-07-25 14:29:22 +03001026 /*
Tejun Heo839a8e82013-04-01 19:08:06 -07001027 * The normal path. Keep writing back @bdi until its
1028 * work_list is empty. Note that this path is also taken
1029 * if @bdi is shutting down even when we're running off the
1030 * rescuer as work_list needs to be drained.
Artem Bityutskiy64677162010-07-25 14:29:22 +03001031 */
Tejun Heo839a8e82013-04-01 19:08:06 -07001032 do {
Wanpeng Li25d130b2013-07-08 16:00:14 -07001033 pages_written = wb_do_writeback(wb);
Tejun Heo839a8e82013-04-01 19:08:06 -07001034 trace_writeback_pages_written(pages_written);
1035 } while (!list_empty(&bdi->work_list));
1036 } else {
1037 /*
1038 * bdi_wq can't get enough workers and we're running off
1039 * the emergency worker. Don't hog it. Hopefully, 1024 is
1040 * enough for efficient IO.
1041 */
1042 pages_written = writeback_inodes_wb(&bdi->wb, 1024,
1043 WB_REASON_FORKER_THREAD);
Dave Chinner455b2862010-07-07 13:24:06 +10001044 trace_writeback_pages_written(pages_written);
Jens Axboe03ba3782009-09-09 09:08:54 +02001045 }
1046
Tejun Heo839a8e82013-04-01 19:08:06 -07001047 if (!list_empty(&bdi->work_list) ||
1048 (wb_has_dirty_io(wb) && dirty_writeback_interval))
1049 queue_delayed_work(bdi_wq, &wb->dwork,
1050 msecs_to_jiffies(dirty_writeback_interval * 10));
Dave Chinner455b2862010-07-07 13:24:06 +10001051
Tejun Heo839a8e82013-04-01 19:08:06 -07001052 current->flags &= ~PF_SWAPWRITE;
Jens Axboe03ba3782009-09-09 09:08:54 +02001053}
1054
1055/*
Jens Axboe03ba3782009-09-09 09:08:54 +02001056 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1057 * the whole world.
1058 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001059void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
Jens Axboe03ba3782009-09-09 09:08:54 +02001060{
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001061 struct backing_dev_info *bdi;
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001062
Jan Kara47df3dd2013-09-11 14:22:22 -07001063 if (!nr_pages)
1064 nr_pages = get_nr_dirty_pages();
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001065
1066 rcu_read_lock();
1067 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1068 if (!bdi_has_dirty_io(bdi))
1069 continue;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001070 __bdi_start_writeback(bdi, nr_pages, false, reason);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001071 }
1072 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +02001073}
1074
1075static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1076{
1077 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1078 struct dentry *dentry;
1079 const char *name = "?";
1080
1081 dentry = d_find_alias(inode);
1082 if (dentry) {
1083 spin_lock(&dentry->d_lock);
1084 name = (const char *) dentry->d_name.name;
1085 }
1086 printk(KERN_DEBUG
1087 "%s(%d): dirtied inode %lu (%s) on %s\n",
1088 current->comm, task_pid_nr(current), inode->i_ino,
1089 name, inode->i_sb->s_id);
1090 if (dentry) {
1091 spin_unlock(&dentry->d_lock);
1092 dput(dentry);
1093 }
1094 }
1095}
1096
1097/**
1098 * __mark_inode_dirty - internal function
1099 * @inode: inode to mark
1100 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1101 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1102 * mark_inode_dirty_sync.
1103 *
1104 * Put the inode on the super block's dirty list.
1105 *
1106 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1107 * dirty list only if it is hashed or if it refers to a blockdev.
1108 * If it was not hashed, it will never be added to the dirty list
1109 * even if it is later hashed, as it will have been marked dirty already.
1110 *
1111 * In short, make sure you hash any inodes _before_ you start marking
1112 * them dirty.
1113 *
Jens Axboe03ba3782009-09-09 09:08:54 +02001114 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1115 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1116 * the kernel-internal blockdev inode represents the dirtying time of the
1117 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1118 * page->mapping->host, so the page-dirtying time is recorded in the internal
1119 * blockdev inode.
1120 */
1121void __mark_inode_dirty(struct inode *inode, int flags)
1122{
1123 struct super_block *sb = inode->i_sb;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001124 struct backing_dev_info *bdi = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +02001125
1126 /*
1127 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1128 * dirty the inode itself
1129 */
1130 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Tejun Heo9fb0a7d2013-01-11 13:06:37 -08001131 trace_writeback_dirty_inode_start(inode, flags);
1132
Jens Axboe03ba3782009-09-09 09:08:54 +02001133 if (sb->s_op->dirty_inode)
Christoph Hellwigaa385722011-05-27 06:53:02 -04001134 sb->s_op->dirty_inode(inode, flags);
Tejun Heo9fb0a7d2013-01-11 13:06:37 -08001135
1136 trace_writeback_dirty_inode(inode, flags);
Jens Axboe03ba3782009-09-09 09:08:54 +02001137 }
1138
1139 /*
1140 * make sure that changes are seen by all cpus before we test i_state
1141 * -- mikulas
1142 */
1143 smp_mb();
1144
1145 /* avoid the locking if we can */
1146 if ((inode->i_state & flags) == flags)
1147 return;
1148
1149 if (unlikely(block_dump))
1150 block_dump___mark_inode_dirty(inode);
1151
Dave Chinner250df6e2011-03-22 22:23:36 +11001152 spin_lock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001153 if ((inode->i_state & flags) != flags) {
1154 const int was_dirty = inode->i_state & I_DIRTY;
1155
1156 inode->i_state |= flags;
1157
1158 /*
1159 * If the inode is being synced, just update its dirty state.
1160 * The unlocker will place the inode on the appropriate
1161 * superblock list, based upon its state.
1162 */
1163 if (inode->i_state & I_SYNC)
Dave Chinner250df6e2011-03-22 22:23:36 +11001164 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001165
1166 /*
1167 * Only add valid (hashed) inodes to the superblock's
1168 * dirty list. Add blockdev inodes as well.
1169 */
1170 if (!S_ISBLK(inode->i_mode)) {
Al Viro1d3382cb2010-10-23 15:19:20 -04001171 if (inode_unhashed(inode))
Dave Chinner250df6e2011-03-22 22:23:36 +11001172 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001173 }
Al Viroa4ffdde2010-06-02 17:38:30 -04001174 if (inode->i_state & I_FREEING)
Dave Chinner250df6e2011-03-22 22:23:36 +11001175 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001176
1177 /*
1178 * If the inode was already on b_dirty/b_io/b_more_io, don't
1179 * reposition it (that would break b_dirty time-ordering).
1180 */
1181 if (!was_dirty) {
Dave Chinnera66979a2011-03-22 22:23:41 +11001182 bool wakeup_bdi = false;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001183 bdi = inode_to_bdi(inode);
Jens Axboe500b0672009-09-09 09:10:25 +02001184
Junxiao Bi146d7002013-09-11 14:23:04 -07001185 spin_unlock(&inode->i_lock);
1186 spin_lock(&bdi->wb.list_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001187 if (bdi_cap_writeback_dirty(bdi)) {
1188 WARN(!test_bit(BDI_registered, &bdi->state),
1189 "bdi-%s not registered\n", bdi->name);
1190
1191 /*
1192 * If this is the first dirty inode for this
1193 * bdi, we have to wake-up the corresponding
1194 * bdi thread to make sure background
1195 * write-back happens later.
1196 */
1197 if (!wb_has_dirty_io(&bdi->wb))
1198 wakeup_bdi = true;
Jens Axboe500b0672009-09-09 09:10:25 +02001199 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001200
1201 inode->dirtied_when = jiffies;
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001202 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001203 spin_unlock(&bdi->wb.list_lock);
Dave Chinnera66979a2011-03-22 22:23:41 +11001204
1205 if (wakeup_bdi)
1206 bdi_wakeup_thread_delayed(bdi);
1207 return;
Jens Axboe03ba3782009-09-09 09:08:54 +02001208 }
1209 }
Dave Chinner250df6e2011-03-22 22:23:36 +11001210out_unlock_inode:
1211 spin_unlock(&inode->i_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001212
Jens Axboe03ba3782009-09-09 09:08:54 +02001213}
1214EXPORT_SYMBOL(__mark_inode_dirty);
1215
Jens Axboeb6e51312009-09-16 15:13:54 +02001216static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001217{
Jens Axboe03ba3782009-09-09 09:08:54 +02001218 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001219
Jens Axboe03ba3782009-09-09 09:08:54 +02001220 /*
1221 * We need to be protected against the filesystem going from
1222 * r/o to r/w or vice versa.
1223 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001224 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001225
Dave Chinner55fa6092011-03-22 22:23:40 +11001226 spin_lock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001227
1228 /*
1229 * Data integrity sync. Must wait for all pages under writeback,
1230 * because there may have been pages dirtied before our sync
1231 * call, but which had writeout started before we write it out.
1232 * In which case, the inode may not be on the dirty list, but
1233 * we still have to wait for that writeout.
1234 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001235 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001236 struct address_space *mapping = inode->i_mapping;
Jens Axboe03ba3782009-09-09 09:08:54 +02001237
Dave Chinner250df6e2011-03-22 22:23:36 +11001238 spin_lock(&inode->i_lock);
1239 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1240 (mapping->nrpages == 0)) {
1241 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001242 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001243 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001244 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001245 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001246 spin_unlock(&inode_sb_list_lock);
1247
Jens Axboe03ba3782009-09-09 09:08:54 +02001248 /*
Dave Chinner55fa6092011-03-22 22:23:40 +11001249 * We hold a reference to 'inode' so it couldn't have been
1250 * removed from s_inodes list while we dropped the
1251 * inode_sb_list_lock. We cannot iput the inode now as we can
1252 * be holding the last reference and we cannot iput it under
1253 * inode_sb_list_lock. So we keep the reference and iput it
1254 * later.
Jens Axboe03ba3782009-09-09 09:08:54 +02001255 */
1256 iput(old_inode);
1257 old_inode = inode;
1258
1259 filemap_fdatawait(mapping);
1260
1261 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001262
Dave Chinner55fa6092011-03-22 22:23:40 +11001263 spin_lock(&inode_sb_list_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001264 }
Dave Chinner55fa6092011-03-22 22:23:40 +11001265 spin_unlock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001266 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267}
1268
Jens Axboed8a85592009-09-02 12:34:32 +02001269/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001270 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
Jens Axboed8a85592009-09-02 12:34:32 +02001271 * @sb: the superblock
Chris Mason3259f8b2010-10-29 11:16:17 -04001272 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001273 * @reason: reason why some writeback work initiated
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 *
Jens Axboed8a85592009-09-02 12:34:32 +02001275 * Start writeback on some inodes on this super_block. No guarantees are made
1276 * on how many (if any) will be written, and this function does not wait
Chris Mason3259f8b2010-10-29 11:16:17 -04001277 * for IO completion of submitted IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001279void writeback_inodes_sb_nr(struct super_block *sb,
1280 unsigned long nr,
1281 enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001283 DECLARE_COMPLETION_ONSTACK(done);
1284 struct wb_writeback_work work = {
Wu Fengguang6e6938b2010-06-06 10:38:15 -06001285 .sb = sb,
1286 .sync_mode = WB_SYNC_NONE,
1287 .tagged_writepages = 1,
1288 .done = &done,
1289 .nr_pages = nr,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001290 .reason = reason,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001291 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001292
Jan Kara6eedc702012-07-03 16:45:27 +02001293 if (sb->s_bdi == &noop_backing_dev_info)
1294 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001295 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001296 bdi_queue_work(sb->s_bdi, &work);
1297 wait_for_completion(&done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298}
Chris Mason3259f8b2010-10-29 11:16:17 -04001299EXPORT_SYMBOL(writeback_inodes_sb_nr);
1300
1301/**
1302 * writeback_inodes_sb - writeback dirty inodes from given super_block
1303 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001304 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001305 *
1306 * Start writeback on some inodes on this super_block. No guarantees are made
1307 * on how many (if any) will be written, and this function does not wait
1308 * for IO completion of submitted IO.
1309 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001310void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001311{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001312 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001313}
Jens Axboed8a85592009-09-02 12:34:32 +02001314EXPORT_SYMBOL(writeback_inodes_sb);
1315
1316/**
Miao Xie10ee27a2013-01-10 13:47:57 +08001317 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
Chris Mason3259f8b2010-10-29 11:16:17 -04001318 * @sb: the superblock
1319 * @nr: the number of pages to write
Miao Xie10ee27a2013-01-10 13:47:57 +08001320 * @reason: the reason of writeback
Chris Mason3259f8b2010-10-29 11:16:17 -04001321 *
Miao Xie10ee27a2013-01-10 13:47:57 +08001322 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
Chris Mason3259f8b2010-10-29 11:16:17 -04001323 * Returns 1 if writeback was started, 0 if not.
1324 */
Miao Xie10ee27a2013-01-10 13:47:57 +08001325int try_to_writeback_inodes_sb_nr(struct super_block *sb,
1326 unsigned long nr,
1327 enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001328{
Miao Xie10ee27a2013-01-10 13:47:57 +08001329 if (writeback_in_progress(sb->s_bdi))
Chris Mason3259f8b2010-10-29 11:16:17 -04001330 return 1;
Miao Xie10ee27a2013-01-10 13:47:57 +08001331
1332 if (!down_read_trylock(&sb->s_umount))
Chris Mason3259f8b2010-10-29 11:16:17 -04001333 return 0;
Miao Xie10ee27a2013-01-10 13:47:57 +08001334
1335 writeback_inodes_sb_nr(sb, nr, reason);
1336 up_read(&sb->s_umount);
1337 return 1;
Chris Mason3259f8b2010-10-29 11:16:17 -04001338}
Miao Xie10ee27a2013-01-10 13:47:57 +08001339EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr);
1340
1341/**
1342 * try_to_writeback_inodes_sb - try to start writeback if none underway
1343 * @sb: the superblock
1344 * @reason: reason why some writeback work was initiated
1345 *
1346 * Implement by try_to_writeback_inodes_sb_nr()
1347 * Returns 1 if writeback was started, 0 if not.
1348 */
1349int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1350{
1351 return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1352}
1353EXPORT_SYMBOL(try_to_writeback_inodes_sb);
Chris Mason3259f8b2010-10-29 11:16:17 -04001354
1355/**
Jens Axboed8a85592009-09-02 12:34:32 +02001356 * sync_inodes_sb - sync sb inode pages
Jan Karac4a391b2013-11-12 15:07:51 -08001357 * @sb: the superblock
1358 * @older_than_this: timestamp
Jens Axboed8a85592009-09-02 12:34:32 +02001359 *
1360 * This function writes and waits on any dirty inode belonging to this
Jan Karac4a391b2013-11-12 15:07:51 -08001361 * superblock that has been dirtied before given timestamp.
Jens Axboed8a85592009-09-02 12:34:32 +02001362 */
Jan Karac4a391b2013-11-12 15:07:51 -08001363void sync_inodes_sb(struct super_block *sb, unsigned long older_than_this)
Jens Axboed8a85592009-09-02 12:34:32 +02001364{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001365 DECLARE_COMPLETION_ONSTACK(done);
1366 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001367 .sb = sb,
1368 .sync_mode = WB_SYNC_ALL,
1369 .nr_pages = LONG_MAX,
Jan Karac4a391b2013-11-12 15:07:51 -08001370 .older_than_this = older_than_this,
1371 .older_than_this_is_set = 1,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001372 .range_cyclic = 0,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001373 .done = &done,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001374 .reason = WB_REASON_SYNC,
Dave Chinner7747bd42013-07-02 22:38:35 +10001375 .for_sync = 1,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001376 };
1377
Jan Kara6eedc702012-07-03 16:45:27 +02001378 /* Nothing to do? */
1379 if (sb->s_bdi == &noop_backing_dev_info)
1380 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001381 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1382
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001383 bdi_queue_work(sb->s_bdi, &work);
1384 wait_for_completion(&done);
1385
Jens Axboeb6e51312009-09-16 15:13:54 +02001386 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001387}
1388EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001391 * write_inode_now - write an inode to disk
1392 * @inode: inode to write to disk
1393 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001395 * This function commits an inode to disk immediately if it is dirty. This is
1396 * primarily needed by knfsd.
1397 *
1398 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400int write_inode_now(struct inode *inode, int sync)
1401{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001402 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 struct writeback_control wbc = {
1404 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001405 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001406 .range_start = 0,
1407 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 };
1409
1410 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001411 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412
1413 might_sleep();
Jan Kara4f8ad652012-05-03 14:48:00 +02001414 return writeback_single_inode(inode, wb, &wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415}
1416EXPORT_SYMBOL(write_inode_now);
1417
1418/**
1419 * sync_inode - write an inode and its pages to disk.
1420 * @inode: the inode to sync
1421 * @wbc: controls the writeback mode
1422 *
1423 * sync_inode() will write an inode and its pages to disk. It will also
1424 * correctly update the inode on its superblock's dirty inode lists and will
1425 * update inode->i_state.
1426 *
1427 * The caller must have a ref on the inode.
1428 */
1429int sync_inode(struct inode *inode, struct writeback_control *wbc)
1430{
Jan Kara4f8ad652012-05-03 14:48:00 +02001431 return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432}
1433EXPORT_SYMBOL(sync_inode);
Christoph Hellwigc37650162010-10-06 10:48:20 +02001434
1435/**
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001436 * sync_inode_metadata - write an inode to disk
Christoph Hellwigc37650162010-10-06 10:48:20 +02001437 * @inode: the inode to sync
1438 * @wait: wait for I/O to complete.
1439 *
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001440 * Write an inode to disk and adjust its dirty state after completion.
Christoph Hellwigc37650162010-10-06 10:48:20 +02001441 *
1442 * Note: only writes the actual inode, no associated data or other metadata.
1443 */
1444int sync_inode_metadata(struct inode *inode, int wait)
1445{
1446 struct writeback_control wbc = {
1447 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1448 .nr_to_write = 0, /* metadata-only */
1449 };
1450
1451 return sync_inode(inode, &wbc);
1452}
1453EXPORT_SYMBOL(sync_inode_metadata);