blob: f855916657ba910f676eeb72ad795317c62fcbd5 [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>
Jens Axboef5ff8422007-09-21 09:19:54 +020017#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/spinlock.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/sched.h>
21#include <linux/fs.h>
22#include <linux/mm.h>
Wu Fengguangbc31b862012-01-07 20:41:55 -060023#include <linux/pagemap.h>
Jens Axboe03ba3782009-09-09 09:08:54 +020024#include <linux/kthread.h>
25#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/writeback.h>
27#include <linux/blkdev.h>
28#include <linux/backing-dev.h>
Dave Chinner455b2862010-07-07 13:24:06 +100029#include <linux/tracepoint.h>
David Howells07f3f052006-09-30 20:52:18 +020030#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Jens Axboed0bceac2009-05-18 08:20:32 +020032/*
Wu Fengguangbc31b862012-01-07 20:41:55 -060033 * 4MB minimal write chunk size
34 */
35#define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10))
36
37/*
Jens Axboec4a77a62009-09-16 15:18:25 +020038 * Passed into wb_writeback(), essentially a subset of writeback_control
39 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020040struct wb_writeback_work {
Jens Axboec4a77a62009-09-16 15:18:25 +020041 long nr_pages;
42 struct super_block *sb;
Wu Fengguangd46db3d2011-05-04 19:54:37 -060043 unsigned long *older_than_this;
Jens Axboec4a77a62009-09-16 15:18:25 +020044 enum writeback_sync_modes sync_mode;
Wu Fengguang6e6938b2010-06-06 10:38:15 -060045 unsigned int tagged_writepages:1;
H Hartley Sweeten52957fe2010-04-01 20:36:30 -050046 unsigned int for_kupdate:1;
47 unsigned int range_cyclic:1;
48 unsigned int for_background:1;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -060049 enum wb_reason reason; /* why was writeback initiated? */
Jens Axboec4a77a62009-09-16 15:18:25 +020050
Jens Axboe8010c3b2009-09-15 20:04:57 +020051 struct list_head list; /* pending work list */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020052 struct completion *done; /* set if the caller waits */
Jens Axboe03ba3782009-09-09 09:08:54 +020053};
54
Dave Chinner455b2862010-07-07 13:24:06 +100055/*
56 * Include the creation of the trace points after defining the
57 * wb_writeback_work structure so that the definition remains local to this
58 * file.
59 */
60#define CREATE_TRACE_POINTS
61#include <trace/events/writeback.h>
62
Dave Chinner455b2862010-07-07 13:24:06 +100063/*
64 * We don't actually have pdflush, but this one is exported though /proc...
65 */
66int nr_pdflush_threads;
67
Adrian Bunkf11b00f2008-04-29 00:58:56 -070068/**
69 * writeback_in_progress - determine whether there is writeback in progress
70 * @bdi: the device's backing_dev_info structure.
71 *
Jens Axboe03ba3782009-09-09 09:08:54 +020072 * Determine whether there is writeback waiting to be handled against a
73 * backing device.
Adrian Bunkf11b00f2008-04-29 00:58:56 -070074 */
75int writeback_in_progress(struct backing_dev_info *bdi)
76{
Jan Kara81d73a32010-08-11 14:17:44 -070077 return test_bit(BDI_writeback_running, &bdi->state);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070078}
79
Jan Kara692ebd12010-09-21 11:51:01 +020080static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
81{
82 struct super_block *sb = inode->i_sb;
Jan Kara692ebd12010-09-21 11:51:01 +020083
Christoph Hellwigaaead252010-10-04 14:25:33 +020084 if (strcmp(sb->s_type->name, "bdev") == 0)
85 return inode->i_mapping->backing_dev_info;
86
87 return sb->s_bdi;
Jan Kara692ebd12010-09-21 11:51:01 +020088}
89
Nick Piggin7ccf19a2010-10-21 11:49:30 +110090static inline struct inode *wb_inode(struct list_head *head)
91{
92 return list_entry(head, struct inode, i_wb_list);
93}
94
Jan Kara65850272011-01-13 15:45:44 -080095/* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
96static void bdi_wakeup_flusher(struct backing_dev_info *bdi)
Nick Piggin4195f732009-05-28 09:01:15 +020097{
Artem Bityutskiyfff5b852010-07-25 14:29:20 +030098 if (bdi->wb.task) {
99 wake_up_process(bdi->wb.task);
100 } else {
101 /*
102 * The bdi thread isn't there, wake up the forker thread which
103 * will create and run it.
104 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200105 wake_up_process(default_backing_dev_info.wb.task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 }
Jan Kara65850272011-01-13 15:45:44 -0800107}
108
109static void bdi_queue_work(struct backing_dev_info *bdi,
110 struct wb_writeback_work *work)
111{
112 trace_writeback_queue(bdi, work);
113
114 spin_lock_bh(&bdi->wb_lock);
115 list_add_tail(&work->list, &bdi->work_list);
116 if (!bdi->wb.task)
117 trace_writeback_nothread(bdi, work);
118 bdi_wakeup_flusher(bdi);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300119 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120}
121
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200122static void
123__bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600124 bool range_cyclic, enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200126 struct wb_writeback_work *work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200127
Jens Axboebcddc3f2009-09-13 20:07:36 +0200128 /*
129 * This is WB_SYNC_NONE writeback, so if allocation fails just
130 * wakeup the thread for old dirty data writeback
131 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200132 work = kzalloc(sizeof(*work), GFP_ATOMIC);
133 if (!work) {
Dave Chinner455b2862010-07-07 13:24:06 +1000134 if (bdi->wb.task) {
135 trace_writeback_nowork(bdi);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200136 wake_up_process(bdi->wb.task);
Dave Chinner455b2862010-07-07 13:24:06 +1000137 }
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200138 return;
Jens Axboebcddc3f2009-09-13 20:07:36 +0200139 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200140
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200141 work->sync_mode = WB_SYNC_NONE;
142 work->nr_pages = nr_pages;
143 work->range_cyclic = range_cyclic;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600144 work->reason = reason;
Christoph Hellwigf0fad8a2009-09-11 09:47:56 +0200145
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200146 bdi_queue_work(bdi, work);
Jens Axboeb6e51312009-09-16 15:13:54 +0200147}
148
149/**
150 * bdi_start_writeback - start writeback
151 * @bdi: the backing device to write from
152 * @nr_pages: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +0800153 * @reason: reason why some writeback work was initiated
Jens Axboeb6e51312009-09-16 15:13:54 +0200154 *
155 * Description:
156 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300157 * started when this function returns, we make no guarantees on
Jens Axboe0e3c9a22010-06-01 11:08:43 +0200158 * completion. Caller need not hold sb s_umount semaphore.
Jens Axboeb6e51312009-09-16 15:13:54 +0200159 *
160 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600161void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
162 enum wb_reason reason)
Jens Axboeb6e51312009-09-16 15:13:54 +0200163{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600164 __bdi_start_writeback(bdi, nr_pages, true, reason);
Christoph Hellwigc5444192010-06-08 18:15:15 +0200165}
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800166
Christoph Hellwigc5444192010-06-08 18:15:15 +0200167/**
168 * bdi_start_background_writeback - start background writeback
169 * @bdi: the backing device to write from
170 *
171 * Description:
Jan Kara65850272011-01-13 15:45:44 -0800172 * This makes sure WB_SYNC_NONE background writeback happens. When
173 * this function returns, it is only guaranteed that for given BDI
174 * some IO is happening if we are over background dirty threshold.
175 * Caller need not hold sb s_umount semaphore.
Christoph Hellwigc5444192010-06-08 18:15:15 +0200176 */
177void bdi_start_background_writeback(struct backing_dev_info *bdi)
178{
Jan Kara65850272011-01-13 15:45:44 -0800179 /*
180 * We just wake up the flusher thread. It will perform background
181 * writeback as soon as there is no other work to do.
182 */
Wu Fengguang71927e82011-01-13 15:45:46 -0800183 trace_writeback_wake_background(bdi);
Jan Kara65850272011-01-13 15:45:44 -0800184 spin_lock_bh(&bdi->wb_lock);
185 bdi_wakeup_flusher(bdi);
186 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187}
188
189/*
Dave Chinnera66979a2011-03-22 22:23:41 +1100190 * Remove the inode from the writeback list it is on.
191 */
192void inode_wb_list_del(struct inode *inode)
193{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600194 struct backing_dev_info *bdi = inode_to_bdi(inode);
Dave Chinnera66979a2011-03-22 22:23:41 +1100195
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600196 spin_lock(&bdi->wb.list_lock);
197 list_del_init(&inode->i_wb_list);
198 spin_unlock(&bdi->wb.list_lock);
199}
Dave Chinnera66979a2011-03-22 22:23:41 +1100200
201/*
Andrew Morton6610a0b2007-10-16 23:30:32 -0700202 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
203 * furthest end of its superblock's dirty-inode list.
204 *
205 * Before stamping the inode's ->dirtied_when, we check to see whether it is
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200206 * already the most-recently-dirtied inode on the b_dirty list. If that is
Andrew Morton6610a0b2007-10-16 23:30:32 -0700207 * the case then the inode must have been redirtied while it was being written
208 * out and we don't reset its dirtied_when.
209 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600210static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
Andrew Morton6610a0b2007-10-16 23:30:32 -0700211{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600212 assert_spin_locked(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200213 if (!list_empty(&wb->b_dirty)) {
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200214 struct inode *tail;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700215
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100216 tail = wb_inode(wb->b_dirty.next);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200217 if (time_before(inode->dirtied_when, tail->dirtied_when))
Andrew Morton6610a0b2007-10-16 23:30:32 -0700218 inode->dirtied_when = jiffies;
219 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100220 list_move(&inode->i_wb_list, &wb->b_dirty);
Andrew Morton6610a0b2007-10-16 23:30:32 -0700221}
222
223/*
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200224 * requeue inode for re-scanning after bdi->b_io list is exhausted.
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700225 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600226static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700227{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600228 assert_spin_locked(&wb->list_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100229 list_move(&inode->i_wb_list, &wb->b_more_io);
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700230}
231
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700232static void inode_sync_complete(struct inode *inode)
233{
234 /*
Dave Chinnera66979a2011-03-22 22:23:41 +1100235 * Prevent speculative execution through
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600236 * spin_unlock(&wb->list_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700237 */
Dave Chinnera66979a2011-03-22 22:23:41 +1100238
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700239 smp_mb();
240 wake_up_bit(&inode->i_state, __I_SYNC);
241}
242
Jeff Laytond2caa3c2009-04-02 16:56:37 -0700243static bool inode_dirtied_after(struct inode *inode, unsigned long t)
244{
245 bool ret = time_after(inode->dirtied_when, t);
246#ifndef CONFIG_64BIT
247 /*
248 * For inodes being constantly redirtied, dirtied_when can get stuck.
249 * It _appears_ to be in the future, but is actually in distant past.
250 * This test is necessary to prevent such wrapped-around relative times
Jens Axboe5b0830c2009-09-23 19:37:09 +0200251 * from permanently stopping the whole bdi writeback.
Jeff Laytond2caa3c2009-04-02 16:56:37 -0700252 */
253 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
254#endif
255 return ret;
256}
257
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700258/*
Fengguang Wu2c136572007-10-16 23:30:39 -0700259 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
260 */
Wu Fengguange84d0a42011-04-23 12:27:27 -0600261static int move_expired_inodes(struct list_head *delaying_queue,
Fengguang Wu2c136572007-10-16 23:30:39 -0700262 struct list_head *dispatch_queue,
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600263 struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700264{
Shaohua Li5c034492009-09-24 14:42:33 +0200265 LIST_HEAD(tmp);
266 struct list_head *pos, *node;
Jens Axboecf137302009-09-24 15:12:57 +0200267 struct super_block *sb = NULL;
Shaohua Li5c034492009-09-24 14:42:33 +0200268 struct inode *inode;
Jens Axboecf137302009-09-24 15:12:57 +0200269 int do_sb_sort = 0;
Wu Fengguange84d0a42011-04-23 12:27:27 -0600270 int moved = 0;
Shaohua Li5c034492009-09-24 14:42:33 +0200271
Fengguang Wu2c136572007-10-16 23:30:39 -0700272 while (!list_empty(delaying_queue)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100273 inode = wb_inode(delaying_queue->prev);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600274 if (work->older_than_this &&
275 inode_dirtied_after(inode, *work->older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700276 break;
Jens Axboecf137302009-09-24 15:12:57 +0200277 if (sb && sb != inode->i_sb)
278 do_sb_sort = 1;
279 sb = inode->i_sb;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100280 list_move(&inode->i_wb_list, &tmp);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600281 moved++;
Shaohua Li5c034492009-09-24 14:42:33 +0200282 }
283
Jens Axboecf137302009-09-24 15:12:57 +0200284 /* just one sb in list, splice to dispatch_queue and we're done */
285 if (!do_sb_sort) {
286 list_splice(&tmp, dispatch_queue);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600287 goto out;
Jens Axboecf137302009-09-24 15:12:57 +0200288 }
289
Shaohua Li5c034492009-09-24 14:42:33 +0200290 /* Move inodes from one superblock together */
291 while (!list_empty(&tmp)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100292 sb = wb_inode(tmp.prev)->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200293 list_for_each_prev_safe(pos, node, &tmp) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100294 inode = wb_inode(pos);
Shaohua Li5c034492009-09-24 14:42:33 +0200295 if (inode->i_sb == sb)
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100296 list_move(&inode->i_wb_list, dispatch_queue);
Shaohua Li5c034492009-09-24 14:42:33 +0200297 }
Fengguang Wu2c136572007-10-16 23:30:39 -0700298 }
Wu Fengguange84d0a42011-04-23 12:27:27 -0600299out:
300 return moved;
Fengguang Wu2c136572007-10-16 23:30:39 -0700301}
302
303/*
304 * Queue all expired dirty inodes for io, eldest first.
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700305 * Before
306 * newly dirtied b_dirty b_io b_more_io
307 * =============> gf edc BA
308 * After
309 * newly dirtied b_dirty b_io b_more_io
310 * =============> g fBAedc
311 * |
312 * +--> dequeue for IO
Fengguang Wu2c136572007-10-16 23:30:39 -0700313 */
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600314static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700315{
Wu Fengguange84d0a42011-04-23 12:27:27 -0600316 int moved;
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600317 assert_spin_locked(&wb->list_lock);
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700318 list_splice_init(&wb->b_more_io, &wb->b_io);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600319 moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
320 trace_writeback_queue_io(wb, work, moved);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200321}
322
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100323static int write_inode(struct inode *inode, struct writeback_control *wbc)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200324{
Jens Axboe03ba3782009-09-09 09:08:54 +0200325 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100326 return inode->i_sb->s_op->write_inode(inode, wbc);
Jens Axboe03ba3782009-09-09 09:08:54 +0200327 return 0;
Fengguang Wu2c136572007-10-16 23:30:39 -0700328}
329
330/*
Christoph Hellwig01c03192009-06-08 13:35:40 +0200331 * Wait for writeback on an inode to complete.
332 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600333static void inode_wait_for_writeback(struct inode *inode,
334 struct bdi_writeback *wb)
Christoph Hellwig01c03192009-06-08 13:35:40 +0200335{
336 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
337 wait_queue_head_t *wqh;
338
339 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100340 while (inode->i_state & I_SYNC) {
341 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600342 spin_unlock(&wb->list_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200343 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600344 spin_lock(&wb->list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100345 spin_lock(&inode->i_lock);
Richard Kennedy58a9d3d2010-05-24 14:32:38 -0700346 }
Christoph Hellwig01c03192009-06-08 13:35:40 +0200347}
348
349/*
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600350 * Write out an inode's dirty pages. Called under wb->list_lock and
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100351 * inode->i_lock. Either the caller has an active reference on the inode or
352 * the inode has I_WILL_FREE set.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200353 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 * If `wait' is set, wait on the writeout.
355 *
356 * The whole writeout design is quite complex and fragile. We want to avoid
357 * starvation of particular inodes when others are being redirtied, prevent
358 * livelocks, etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 */
360static int
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600361writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
362 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 struct address_space *mapping = inode->i_mapping;
Wu Fengguang251d6a42010-12-01 17:33:37 -0600365 long nr_to_write = wbc->nr_to_write;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200366 unsigned dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 int ret;
368
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600369 assert_spin_locked(&wb->list_lock);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100370 assert_spin_locked(&inode->i_lock);
371
Christoph Hellwig01c03192009-06-08 13:35:40 +0200372 if (!atomic_read(&inode->i_count))
373 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
374 else
375 WARN_ON(inode->i_state & I_WILL_FREE);
376
377 if (inode->i_state & I_SYNC) {
378 /*
379 * If this inode is locked for writeback and we are not doing
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200380 * writeback-for-data-integrity, move it to b_more_io so that
Christoph Hellwig01c03192009-06-08 13:35:40 +0200381 * writeback can proceed with the other inodes on s_io.
382 *
383 * We'll have another go at writing back this inode when we
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200384 * completed a full scan of b_io.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200385 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100386 if (wbc->sync_mode != WB_SYNC_ALL) {
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600387 requeue_io(inode, wb);
Wu Fengguang251d6a42010-12-01 17:33:37 -0600388 trace_writeback_single_inode_requeue(inode, wbc,
389 nr_to_write);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200390 return 0;
391 }
392
393 /*
394 * It's a data-integrity sync. We must wait.
395 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600396 inode_wait_for_writeback(inode, wb);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200397 }
398
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700399 BUG_ON(inode->i_state & I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400401 /* Set I_SYNC, reset I_DIRTY_PAGES */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700402 inode->i_state |= I_SYNC;
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400403 inode->i_state &= ~I_DIRTY_PAGES;
Dave Chinner250df6e2011-03-22 22:23:36 +1100404 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600405 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
407 ret = do_writepages(mapping, wbc);
408
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100409 /*
410 * Make sure to wait on the data before writing out the metadata.
411 * This is important for filesystems that modify metadata on data
412 * I/O completion.
413 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100414 if (wbc->sync_mode == WB_SYNC_ALL) {
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100415 int err = filemap_fdatawait(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 if (ret == 0)
417 ret = err;
418 }
419
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400420 /*
421 * Some filesystems may redirty the inode during the writeback
422 * due to delalloc, clear dirty metadata flags right before
423 * write_inode()
424 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100425 spin_lock(&inode->i_lock);
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400426 dirty = inode->i_state & I_DIRTY;
427 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100428 spin_unlock(&inode->i_lock);
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100429 /* Don't write the inode if only I_DIRTY_PAGES was set */
430 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100431 int err = write_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 if (ret == 0)
433 ret = err;
434 }
435
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600436 spin_lock(&wb->list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100437 spin_lock(&inode->i_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700438 inode->i_state &= ~I_SYNC;
Al Viroa4ffdde2010-06-02 17:38:30 -0400439 if (!(inode->i_state & I_FREEING)) {
Wu Fengguang94c3dcb2011-04-27 19:05:21 -0600440 /*
441 * Sync livelock prevention. Each inode is tagged and synced in
442 * one shot. If still dirty, it will be redirty_tail()'ed below.
443 * Update the dirty time to prevent enqueue and sync it again.
444 */
445 if ((inode->i_state & I_DIRTY) &&
446 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
447 inode->dirtied_when = jiffies;
448
Wu Fengguang23539af2010-08-11 14:17:41 -0700449 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 /*
451 * We didn't write back all the pages. nfs_writepages()
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700452 * sometimes bales out without doing anything.
Andrew Morton1b43ef92007-10-16 23:30:35 -0700453 */
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700454 inode->i_state |= I_DIRTY_PAGES;
455 if (wbc->nr_to_write <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 /*
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700457 * slice used up: queue for next turn
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600459 requeue_io(inode, wb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 } else {
461 /*
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700462 * Writeback blocked by something other than
463 * congestion. Delay the inode for some time to
464 * avoid spinning on the CPU (100% iowait)
465 * retrying writeback of the dirty page/inode
466 * that cannot be performed immediately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600468 redirty_tail(inode, wb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 }
Wu Fengguang23539af2010-08-11 14:17:41 -0700470 } else if (inode->i_state & I_DIRTY) {
471 /*
472 * Filesystems can dirty the inode during writeback
473 * operations, such as delayed allocation during
474 * submission or metadata updates after data IO
475 * completion.
476 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600477 redirty_tail(inode, wb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 } else {
479 /*
Nick Piggin9e38d862010-10-23 06:55:17 -0400480 * The inode is clean. At this point we either have
481 * a reference to the inode or it's on it's way out.
482 * No need to add it back to the LRU.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 */
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100484 list_del_init(&inode->i_wb_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 }
486 }
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700487 inode_sync_complete(inode);
Wu Fengguang251d6a42010-12-01 17:33:37 -0600488 trace_writeback_single_inode(inode, wbc, nr_to_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 return ret;
490}
491
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600492static long writeback_chunk_size(struct backing_dev_info *bdi,
493 struct wb_writeback_work *work)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600494{
495 long pages;
496
497 /*
498 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
499 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
500 * here avoids calling into writeback_inodes_wb() more than once.
501 *
502 * The intended call sequence for WB_SYNC_ALL writeback is:
503 *
504 * wb_writeback()
505 * writeback_sb_inodes() <== called only once
506 * write_cache_pages() <== called once for each inode
507 * (quickly) tag currently dirty pages
508 * (maybe slowly) sync all tagged pages
509 */
510 if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
511 pages = LONG_MAX;
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600512 else {
513 pages = min(bdi->avg_write_bandwidth / 2,
514 global_dirty_limit / DIRTY_SCOPE);
515 pages = min(pages, work->nr_pages);
516 pages = round_down(pages + MIN_WRITEBACK_PAGES,
517 MIN_WRITEBACK_PAGES);
518 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600519
520 return pages;
521}
522
Jens Axboe03ba3782009-09-09 09:08:54 +0200523/*
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800524 * Write a portion of b_io inodes which belong to @sb.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200525 *
526 * If @only_this_sb is true, then find and write all such
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800527 * inodes. Otherwise write only ones which go sequentially
528 * in reverse order.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200529 *
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600530 * Return the number of pages and/or inodes written.
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800531 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600532static long writeback_sb_inodes(struct super_block *sb,
533 struct bdi_writeback *wb,
534 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200535{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600536 struct writeback_control wbc = {
537 .sync_mode = work->sync_mode,
538 .tagged_writepages = work->tagged_writepages,
539 .for_kupdate = work->for_kupdate,
540 .for_background = work->for_background,
541 .range_cyclic = work->range_cyclic,
542 .range_start = 0,
543 .range_end = LLONG_MAX,
544 };
545 unsigned long start_time = jiffies;
546 long write_chunk;
547 long wrote = 0; /* count both pages and inodes */
548
Jens Axboe03ba3782009-09-09 09:08:54 +0200549 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100550 struct inode *inode = wb_inode(wb->b_io.prev);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200551
552 if (inode->i_sb != sb) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600553 if (work->sb) {
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200554 /*
555 * We only want to write back data for this
556 * superblock, move all inodes not belonging
557 * to it back onto the dirty list.
558 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600559 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200560 continue;
561 }
562
563 /*
564 * The inode belongs to a different superblock.
565 * Bounce back to the caller to unpin this and
566 * pin the next superblock.
567 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600568 break;
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200569 }
570
Christoph Hellwig9843b762010-10-24 19:40:46 +0200571 /*
572 * Don't bother with new inodes or inodes beeing freed, first
573 * kind does not need peridic writeout yet, and for the latter
574 * kind writeout is handled by the freer.
575 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100576 spin_lock(&inode->i_lock);
Christoph Hellwig9843b762010-10-24 19:40:46 +0200577 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100578 spin_unlock(&inode->i_lock);
Wu Fengguangfcc5c222011-07-11 23:08:50 -0700579 redirty_tail(inode, wb);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700580 continue;
581 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 __iget(inode);
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600583 write_chunk = writeback_chunk_size(wb->bdi, work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600584 wbc.nr_to_write = write_chunk;
585 wbc.pages_skipped = 0;
Dave Chinner250df6e2011-03-22 22:23:36 +1100586
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600587 writeback_single_inode(inode, wb, &wbc);
588
589 work->nr_pages -= write_chunk - wbc.nr_to_write;
590 wrote += write_chunk - wbc.nr_to_write;
591 if (!(inode->i_state & I_DIRTY))
592 wrote++;
593 if (wbc.pages_skipped) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 /*
595 * writeback is not making progress due to locked
596 * buffers. Skip this inode for now.
597 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600598 redirty_tail(inode, wb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 }
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100600 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600601 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 iput(inode);
OGAWA Hirofumi4ffc8442006-03-25 03:07:44 -0800603 cond_resched();
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600604 spin_lock(&wb->list_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600605 /*
606 * bail out to wb_writeback() often enough to check
607 * background threshold and other termination conditions.
608 */
609 if (wrote) {
610 if (time_is_before_jiffies(start_time + HZ / 10UL))
611 break;
612 if (work->nr_pages <= 0)
613 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800614 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600616 return wrote;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800617}
Nick Piggin38f21972009-01-06 14:40:25 -0800618
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600619static long __writeback_inodes_wb(struct bdi_writeback *wb,
620 struct wb_writeback_work *work)
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800621{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600622 unsigned long start_time = jiffies;
623 long wrote = 0;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800624
625 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100626 struct inode *inode = wb_inode(wb->b_io.prev);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800627 struct super_block *sb = inode->i_sb;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800628
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000629 if (!grab_super_passive(sb)) {
Wu Fengguang0e995812011-07-29 22:14:35 -0600630 /*
631 * grab_super_passive() may fail consistently due to
632 * s_umount being grabbed by someone else. Don't use
633 * requeue_io() to avoid busy retrying the inode/sb.
634 */
635 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200636 continue;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800637 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600638 wrote += writeback_sb_inodes(sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200639 drop_super(sb);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800640
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600641 /* refer to the same tests at the end of writeback_sb_inodes */
642 if (wrote) {
643 if (time_is_before_jiffies(start_time + HZ / 10UL))
644 break;
645 if (work->nr_pages <= 0)
646 break;
647 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800648 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200649 /* Leave any unwritten inodes on b_io */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600650 return wrote;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200651}
652
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600653long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
654 enum wb_reason reason)
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200655{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600656 struct wb_writeback_work work = {
657 .nr_pages = nr_pages,
658 .sync_mode = WB_SYNC_NONE,
659 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600660 .reason = reason,
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600661 };
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200662
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600663 spin_lock(&wb->list_lock);
Wu Fengguang424b3512010-07-21 20:11:53 -0600664 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600665 queue_io(wb, &work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600666 __writeback_inodes_wb(wb, &work);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600667 spin_unlock(&wb->list_lock);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200668
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600669 return nr_pages - work.nr_pages;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200670}
671
Wu Fengguangb00949a2010-11-18 14:38:33 -0600672static bool over_bground_thresh(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200673{
674 unsigned long background_thresh, dirty_thresh;
675
Wu Fengguang16c40422010-08-11 14:17:39 -0700676 global_dirty_limits(&background_thresh, &dirty_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200677
Wu Fengguangb00949a2010-11-18 14:38:33 -0600678 if (global_page_state(NR_FILE_DIRTY) +
679 global_page_state(NR_UNSTABLE_NFS) > background_thresh)
680 return true;
681
682 if (bdi_stat(bdi, BDI_RECLAIMABLE) >
683 bdi_dirty_limit(bdi, background_thresh))
684 return true;
685
686 return false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200687}
688
689/*
Wu Fengguange98be2d2010-08-29 11:22:30 -0600690 * Called under wb->list_lock. If there are multiple wb per bdi,
691 * only the flusher working on the first wb should do it.
692 */
693static void wb_update_bandwidth(struct bdi_writeback *wb,
694 unsigned long start_time)
695{
Wu Fengguangaf6a3112011-10-03 20:46:17 -0600696 __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
Wu Fengguange98be2d2010-08-29 11:22:30 -0600697}
698
699/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200700 * Explicit flushing or periodic writeback of "old" data.
701 *
702 * Define "old": the first time one of an inode's pages is dirtied, we mark the
703 * dirtying-time in the inode's address_space. So this periodic writeback code
704 * just walks the superblock inode list, writing back any inodes which are
705 * older than a specific point in time.
706 *
707 * Try to run once per dirty_writeback_interval. But if a writeback event
708 * takes longer than a dirty_writeback_interval interval, then leave a
709 * one-second gap.
710 *
711 * older_than_this takes precedence over nr_to_write. So we'll only write back
712 * all dirty pages if they are all attached to "old" mappings.
713 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200714static long wb_writeback(struct bdi_writeback *wb,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200715 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200716{
Wu Fengguange98be2d2010-08-29 11:22:30 -0600717 unsigned long wb_start = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600718 long nr_pages = work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200719 unsigned long oldest_jif;
Jan Karaa5989bd2009-09-16 19:22:48 +0200720 struct inode *inode;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600721 long progress;
Jens Axboe03ba3782009-09-09 09:08:54 +0200722
Wu Fengguange185dda2011-04-23 11:26:07 -0600723 oldest_jif = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600724 work->older_than_this = &oldest_jif;
Jens Axboe03ba3782009-09-09 09:08:54 +0200725
Wu Fengguange8dfc302011-04-21 12:06:32 -0600726 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200727 for (;;) {
728 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800729 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200730 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200731 if (work->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200732 break;
733
734 /*
Jan Karaaa373cf2011-01-13 15:45:47 -0800735 * Background writeout and kupdate-style writeback may
736 * run forever. Stop them if there is other work to do
737 * so that e.g. sync can proceed. They'll be restarted
738 * after the other works are all done.
739 */
740 if ((work->for_background || work->for_kupdate) &&
741 !list_empty(&wb->bdi->work_list))
742 break;
743
744 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800745 * For background writeout, stop when we are below the
746 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200747 */
Wu Fengguangb00949a2010-11-18 14:38:33 -0600748 if (work->for_background && !over_bground_thresh(wb->bdi))
Jens Axboe03ba3782009-09-09 09:08:54 +0200749 break;
750
Jan Kara1bc36b62011-10-19 11:44:41 +0200751 /*
752 * Kupdate and background works are special and we want to
753 * include all inodes that need writing. Livelock avoidance is
754 * handled by these works yielding to any other work so we are
755 * safe.
756 */
Wu Fengguangba9aa832010-07-21 20:32:30 -0600757 if (work->for_kupdate) {
758 oldest_jif = jiffies -
759 msecs_to_jiffies(dirty_expire_interval * 10);
Jan Kara1bc36b62011-10-19 11:44:41 +0200760 } else if (work->for_background)
761 oldest_jif = jiffies;
Dave Chinner028c2dd2010-07-07 13:24:07 +1000762
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600763 trace_writeback_start(wb->bdi, work);
Wu Fengguange8dfc302011-04-21 12:06:32 -0600764 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600765 queue_io(wb, work);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200766 if (work->sb)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600767 progress = writeback_sb_inodes(work->sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200768 else
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600769 progress = __writeback_inodes_wb(wb, work);
770 trace_writeback_written(wb->bdi, work);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000771
Wu Fengguange98be2d2010-08-29 11:22:30 -0600772 wb_update_bandwidth(wb, wb_start);
Jens Axboe03ba3782009-09-09 09:08:54 +0200773
774 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200775 * Did we write something? Try for more
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600776 *
777 * Dirty inodes are moved to b_io for writeback in batches.
778 * The completion of the current batch does not necessarily
779 * mean the overall work is done. So we keep looping as long
780 * as made some progress on cleaning pages or inodes.
Jens Axboe71fd05a2009-09-23 19:32:26 +0200781 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600782 if (progress)
Jens Axboe71fd05a2009-09-23 19:32:26 +0200783 continue;
784 /*
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600785 * No more inodes for IO, bail
Jens Axboe71fd05a2009-09-23 19:32:26 +0200786 */
Wu Fengguangb7a24412010-07-21 22:19:51 -0600787 if (list_empty(&wb->b_more_io))
Jens Axboe03ba3782009-09-09 09:08:54 +0200788 break;
789 /*
Jens Axboe8010c3b2009-09-15 20:04:57 +0200790 * Nothing written. Wait for some inode to
791 * become available for writeback. Otherwise
792 * we'll just busyloop.
Jens Axboe03ba3782009-09-09 09:08:54 +0200793 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200794 if (!list_empty(&wb->b_more_io)) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600795 trace_writeback_wait(wb->bdi, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200796 inode = wb_inode(wb->b_more_io.prev);
Dave Chinner250df6e2011-03-22 22:23:36 +1100797 spin_lock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600798 inode_wait_for_writeback(inode, wb);
Dave Chinner250df6e2011-03-22 22:23:36 +1100799 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200800 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200801 }
Wu Fengguange8dfc302011-04-21 12:06:32 -0600802 spin_unlock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200803
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600804 return nr_pages - work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200805}
806
807/*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200808 * Return the next wb_writeback_work struct that hasn't been processed yet.
Jens Axboe03ba3782009-09-09 09:08:54 +0200809 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200810static struct wb_writeback_work *
Minchan Kim08852b62010-08-03 12:51:16 +0200811get_next_work_item(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200812{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200813 struct wb_writeback_work *work = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200814
Artem Bityutskiy64677162010-07-25 14:29:22 +0300815 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200816 if (!list_empty(&bdi->work_list)) {
817 work = list_entry(bdi->work_list.next,
818 struct wb_writeback_work, list);
819 list_del_init(&work->list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200820 }
Artem Bityutskiy64677162010-07-25 14:29:22 +0300821 spin_unlock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200822 return work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200823}
824
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700825/*
826 * Add in the number of potentially dirty inodes, because each inode
827 * write can dirty pagecache in the underlying blockdev.
828 */
829static unsigned long get_nr_dirty_pages(void)
830{
831 return global_page_state(NR_FILE_DIRTY) +
832 global_page_state(NR_UNSTABLE_NFS) +
833 get_nr_dirty_inodes();
834}
835
Jan Kara65850272011-01-13 15:45:44 -0800836static long wb_check_background_flush(struct bdi_writeback *wb)
837{
Wu Fengguangb00949a2010-11-18 14:38:33 -0600838 if (over_bground_thresh(wb->bdi)) {
Jan Kara65850272011-01-13 15:45:44 -0800839
840 struct wb_writeback_work work = {
841 .nr_pages = LONG_MAX,
842 .sync_mode = WB_SYNC_NONE,
843 .for_background = 1,
844 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600845 .reason = WB_REASON_BACKGROUND,
Jan Kara65850272011-01-13 15:45:44 -0800846 };
847
848 return wb_writeback(wb, &work);
849 }
850
851 return 0;
852}
853
Jens Axboe03ba3782009-09-09 09:08:54 +0200854static long wb_check_old_data_flush(struct bdi_writeback *wb)
855{
856 unsigned long expired;
857 long nr_pages;
858
Jens Axboe69b62d02010-05-17 12:51:03 +0200859 /*
860 * When set to zero, disable periodic writeback
861 */
862 if (!dirty_writeback_interval)
863 return 0;
864
Jens Axboe03ba3782009-09-09 09:08:54 +0200865 expired = wb->last_old_flush +
866 msecs_to_jiffies(dirty_writeback_interval * 10);
867 if (time_before(jiffies, expired))
868 return 0;
869
870 wb->last_old_flush = jiffies;
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700871 nr_pages = get_nr_dirty_pages();
Jens Axboe03ba3782009-09-09 09:08:54 +0200872
Jens Axboec4a77a62009-09-16 15:18:25 +0200873 if (nr_pages) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200874 struct wb_writeback_work work = {
Jens Axboec4a77a62009-09-16 15:18:25 +0200875 .nr_pages = nr_pages,
876 .sync_mode = WB_SYNC_NONE,
877 .for_kupdate = 1,
878 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600879 .reason = WB_REASON_PERIODIC,
Jens Axboec4a77a62009-09-16 15:18:25 +0200880 };
881
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200882 return wb_writeback(wb, &work);
Jens Axboec4a77a62009-09-16 15:18:25 +0200883 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200884
885 return 0;
886}
887
888/*
889 * Retrieve work items and do the writeback they describe
890 */
891long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
892{
893 struct backing_dev_info *bdi = wb->bdi;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200894 struct wb_writeback_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +0200895 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200896
Jan Kara81d73a32010-08-11 14:17:44 -0700897 set_bit(BDI_writeback_running, &wb->bdi->state);
Minchan Kim08852b62010-08-03 12:51:16 +0200898 while ((work = get_next_work_item(bdi)) != NULL) {
Jens Axboe03ba3782009-09-09 09:08:54 +0200899 /*
900 * Override sync mode, in case we must wait for completion
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200901 * because this thread is exiting now.
Jens Axboe03ba3782009-09-09 09:08:54 +0200902 */
903 if (force_wait)
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200904 work->sync_mode = WB_SYNC_ALL;
905
Dave Chinner455b2862010-07-07 13:24:06 +1000906 trace_writeback_exec(bdi, work);
907
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200908 wrote += wb_writeback(wb, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200909
910 /*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200911 * Notify the caller of completion if this is a synchronous
912 * work item, otherwise just free it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200913 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200914 if (work->done)
915 complete(work->done);
916 else
917 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200918 }
919
920 /*
921 * Check for periodic writeback, kupdated() style
922 */
923 wrote += wb_check_old_data_flush(wb);
Jan Kara65850272011-01-13 15:45:44 -0800924 wrote += wb_check_background_flush(wb);
Jan Kara81d73a32010-08-11 14:17:44 -0700925 clear_bit(BDI_writeback_running, &wb->bdi->state);
Jens Axboe03ba3782009-09-09 09:08:54 +0200926
927 return wrote;
928}
929
930/*
931 * Handle writeback of dirty data for the device backed by this bdi. Also
932 * wakes up periodically and does kupdated style flushing.
933 */
Christoph Hellwig08243902010-06-19 23:08:22 +0200934int bdi_writeback_thread(void *data)
Jens Axboe03ba3782009-09-09 09:08:54 +0200935{
Christoph Hellwig08243902010-06-19 23:08:22 +0200936 struct bdi_writeback *wb = data;
937 struct backing_dev_info *bdi = wb->bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +0200938 long pages_written;
939
Peter Zijlstra766f9162010-10-26 14:22:45 -0700940 current->flags |= PF_SWAPWRITE;
Christoph Hellwig08243902010-06-19 23:08:22 +0200941 set_freezable();
Artem Bityutskiyecd58402010-07-25 14:29:18 +0300942 wb->last_active = jiffies;
Christoph Hellwig08243902010-06-19 23:08:22 +0200943
944 /*
945 * Our parent may run at a different priority, just set us to normal
946 */
947 set_user_nice(current, 0);
948
Dave Chinner455b2862010-07-07 13:24:06 +1000949 trace_writeback_thread_start(bdi);
950
Tejun Heo8a32c442011-11-21 12:32:23 -0800951 while (!kthread_freezable_should_stop(NULL)) {
Artem Bityutskiy64677162010-07-25 14:29:22 +0300952 /*
953 * Remove own delayed wake-up timer, since we are already awake
954 * and we'll take care of the preriodic write-back.
955 */
956 del_timer(&wb->wakeup_timer);
957
Jens Axboe03ba3782009-09-09 09:08:54 +0200958 pages_written = wb_do_writeback(wb, 0);
959
Dave Chinner455b2862010-07-07 13:24:06 +1000960 trace_writeback_pages_written(pages_written);
Jens Axboe03ba3782009-09-09 09:08:54 +0200961
Jens Axboe03ba3782009-09-09 09:08:54 +0200962 if (pages_written)
Artem Bityutskiyecd58402010-07-25 14:29:18 +0300963 wb->last_active = jiffies;
Jens Axboe03ba3782009-09-09 09:08:54 +0200964
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300965 set_current_state(TASK_INTERRUPTIBLE);
J. Bruce Fieldsb76b4012010-08-28 08:52:10 +0200966 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300967 __set_current_state(TASK_RUNNING);
968 continue;
Jens Axboe03ba3782009-09-09 09:08:54 +0200969 }
970
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300971 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300972 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300973 else {
974 /*
975 * We have nothing to do, so can go sleep without any
976 * timeout and save power. When a work is queued or
977 * something is made dirty - we will be woken up.
978 */
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300979 schedule();
Jens Axboef9eadbb2010-05-18 14:31:45 +0200980 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200981 }
982
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300983 /* Flush any work that raced with us exiting */
Christoph Hellwig08243902010-06-19 23:08:22 +0200984 if (!list_empty(&bdi->work_list))
985 wb_do_writeback(wb, 1);
Dave Chinner455b2862010-07-07 13:24:06 +1000986
987 trace_writeback_thread_stop(bdi);
Jens Axboe03ba3782009-09-09 09:08:54 +0200988 return 0;
989}
990
Christoph Hellwig08243902010-06-19 23:08:22 +0200991
Jens Axboe03ba3782009-09-09 09:08:54 +0200992/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200993 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
994 * the whole world.
995 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600996void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
Jens Axboe03ba3782009-09-09 09:08:54 +0200997{
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200998 struct backing_dev_info *bdi;
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200999
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001000 if (!nr_pages) {
1001 nr_pages = global_page_state(NR_FILE_DIRTY) +
Jens Axboe03ba3782009-09-09 09:08:54 +02001002 global_page_state(NR_UNSTABLE_NFS);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001003 }
1004
1005 rcu_read_lock();
1006 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1007 if (!bdi_has_dirty_io(bdi))
1008 continue;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001009 __bdi_start_writeback(bdi, nr_pages, false, reason);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001010 }
1011 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +02001012}
1013
1014static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1015{
1016 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1017 struct dentry *dentry;
1018 const char *name = "?";
1019
1020 dentry = d_find_alias(inode);
1021 if (dentry) {
1022 spin_lock(&dentry->d_lock);
1023 name = (const char *) dentry->d_name.name;
1024 }
1025 printk(KERN_DEBUG
1026 "%s(%d): dirtied inode %lu (%s) on %s\n",
1027 current->comm, task_pid_nr(current), inode->i_ino,
1028 name, inode->i_sb->s_id);
1029 if (dentry) {
1030 spin_unlock(&dentry->d_lock);
1031 dput(dentry);
1032 }
1033 }
1034}
1035
1036/**
1037 * __mark_inode_dirty - internal function
1038 * @inode: inode to mark
1039 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1040 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1041 * mark_inode_dirty_sync.
1042 *
1043 * Put the inode on the super block's dirty list.
1044 *
1045 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1046 * dirty list only if it is hashed or if it refers to a blockdev.
1047 * If it was not hashed, it will never be added to the dirty list
1048 * even if it is later hashed, as it will have been marked dirty already.
1049 *
1050 * In short, make sure you hash any inodes _before_ you start marking
1051 * them dirty.
1052 *
Jens Axboe03ba3782009-09-09 09:08:54 +02001053 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1054 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1055 * the kernel-internal blockdev inode represents the dirtying time of the
1056 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1057 * page->mapping->host, so the page-dirtying time is recorded in the internal
1058 * blockdev inode.
1059 */
1060void __mark_inode_dirty(struct inode *inode, int flags)
1061{
1062 struct super_block *sb = inode->i_sb;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001063 struct backing_dev_info *bdi = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +02001064
1065 /*
1066 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1067 * dirty the inode itself
1068 */
1069 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1070 if (sb->s_op->dirty_inode)
Christoph Hellwigaa385722011-05-27 06:53:02 -04001071 sb->s_op->dirty_inode(inode, flags);
Jens Axboe03ba3782009-09-09 09:08:54 +02001072 }
1073
1074 /*
1075 * make sure that changes are seen by all cpus before we test i_state
1076 * -- mikulas
1077 */
1078 smp_mb();
1079
1080 /* avoid the locking if we can */
1081 if ((inode->i_state & flags) == flags)
1082 return;
1083
1084 if (unlikely(block_dump))
1085 block_dump___mark_inode_dirty(inode);
1086
Dave Chinner250df6e2011-03-22 22:23:36 +11001087 spin_lock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001088 if ((inode->i_state & flags) != flags) {
1089 const int was_dirty = inode->i_state & I_DIRTY;
1090
1091 inode->i_state |= flags;
1092
1093 /*
1094 * If the inode is being synced, just update its dirty state.
1095 * The unlocker will place the inode on the appropriate
1096 * superblock list, based upon its state.
1097 */
1098 if (inode->i_state & I_SYNC)
Dave Chinner250df6e2011-03-22 22:23:36 +11001099 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001100
1101 /*
1102 * Only add valid (hashed) inodes to the superblock's
1103 * dirty list. Add blockdev inodes as well.
1104 */
1105 if (!S_ISBLK(inode->i_mode)) {
Al Viro1d3382c2010-10-23 15:19:20 -04001106 if (inode_unhashed(inode))
Dave Chinner250df6e2011-03-22 22:23:36 +11001107 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001108 }
Al Viroa4ffdde2010-06-02 17:38:30 -04001109 if (inode->i_state & I_FREEING)
Dave Chinner250df6e2011-03-22 22:23:36 +11001110 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001111
1112 /*
1113 * If the inode was already on b_dirty/b_io/b_more_io, don't
1114 * reposition it (that would break b_dirty time-ordering).
1115 */
1116 if (!was_dirty) {
Dave Chinnera66979a2011-03-22 22:23:41 +11001117 bool wakeup_bdi = false;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001118 bdi = inode_to_bdi(inode);
Jens Axboe500b0672009-09-09 09:10:25 +02001119
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001120 if (bdi_cap_writeback_dirty(bdi)) {
1121 WARN(!test_bit(BDI_registered, &bdi->state),
1122 "bdi-%s not registered\n", bdi->name);
1123
1124 /*
1125 * If this is the first dirty inode for this
1126 * bdi, we have to wake-up the corresponding
1127 * bdi thread to make sure background
1128 * write-back happens later.
1129 */
1130 if (!wb_has_dirty_io(&bdi->wb))
1131 wakeup_bdi = true;
Jens Axboe500b0672009-09-09 09:10:25 +02001132 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001133
Dave Chinnera66979a2011-03-22 22:23:41 +11001134 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001135 spin_lock(&bdi->wb.list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001136 inode->dirtied_when = jiffies;
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001137 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001138 spin_unlock(&bdi->wb.list_lock);
Dave Chinnera66979a2011-03-22 22:23:41 +11001139
1140 if (wakeup_bdi)
1141 bdi_wakeup_thread_delayed(bdi);
1142 return;
Jens Axboe03ba3782009-09-09 09:08:54 +02001143 }
1144 }
Dave Chinner250df6e2011-03-22 22:23:36 +11001145out_unlock_inode:
1146 spin_unlock(&inode->i_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001147
Jens Axboe03ba3782009-09-09 09:08:54 +02001148}
1149EXPORT_SYMBOL(__mark_inode_dirty);
1150
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001151/*
1152 * Write out a superblock's list of dirty inodes. A wait will be performed
1153 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1154 *
1155 * If older_than_this is non-NULL, then only write out inodes which
1156 * had their first dirtying at a time earlier than *older_than_this.
1157 *
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001158 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1159 * This function assumes that the blockdev superblock's inodes are backed by
1160 * a variety of queues, so all inodes are searched. For other superblocks,
1161 * assume that all inodes are backed by the same queue.
1162 *
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001163 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1164 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1165 * on the writer throttling path, and we get decent balancing between many
1166 * throttled threads: we don't want them all piling up on inode_sync_wait.
1167 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001168static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001169{
Jens Axboe03ba3782009-09-09 09:08:54 +02001170 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001171
Jens Axboe03ba3782009-09-09 09:08:54 +02001172 /*
1173 * We need to be protected against the filesystem going from
1174 * r/o to r/w or vice versa.
1175 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001176 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001177
Dave Chinner55fa6092011-03-22 22:23:40 +11001178 spin_lock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001179
1180 /*
1181 * Data integrity sync. Must wait for all pages under writeback,
1182 * because there may have been pages dirtied before our sync
1183 * call, but which had writeout started before we write it out.
1184 * In which case, the inode may not be on the dirty list, but
1185 * we still have to wait for that writeout.
1186 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001187 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001188 struct address_space *mapping = inode->i_mapping;
Jens Axboe03ba3782009-09-09 09:08:54 +02001189
Dave Chinner250df6e2011-03-22 22:23:36 +11001190 spin_lock(&inode->i_lock);
1191 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1192 (mapping->nrpages == 0)) {
1193 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001194 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001195 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001196 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001197 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001198 spin_unlock(&inode_sb_list_lock);
1199
Jens Axboe03ba3782009-09-09 09:08:54 +02001200 /*
Dave Chinner55fa6092011-03-22 22:23:40 +11001201 * We hold a reference to 'inode' so it couldn't have been
1202 * removed from s_inodes list while we dropped the
1203 * inode_sb_list_lock. We cannot iput the inode now as we can
1204 * be holding the last reference and we cannot iput it under
1205 * inode_sb_list_lock. So we keep the reference and iput it
1206 * later.
Jens Axboe03ba3782009-09-09 09:08:54 +02001207 */
1208 iput(old_inode);
1209 old_inode = inode;
1210
1211 filemap_fdatawait(mapping);
1212
1213 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001214
Dave Chinner55fa6092011-03-22 22:23:40 +11001215 spin_lock(&inode_sb_list_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001216 }
Dave Chinner55fa6092011-03-22 22:23:40 +11001217 spin_unlock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001218 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219}
1220
Jens Axboed8a85592009-09-02 12:34:32 +02001221/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001222 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
Jens Axboed8a85592009-09-02 12:34:32 +02001223 * @sb: the superblock
Chris Mason3259f8b2010-10-29 11:16:17 -04001224 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001225 * @reason: reason why some writeback work initiated
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 *
Jens Axboed8a85592009-09-02 12:34:32 +02001227 * Start writeback on some inodes on this super_block. No guarantees are made
1228 * on how many (if any) will be written, and this function does not wait
Chris Mason3259f8b2010-10-29 11:16:17 -04001229 * for IO completion of submitted IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001231void writeback_inodes_sb_nr(struct super_block *sb,
1232 unsigned long nr,
1233 enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001235 DECLARE_COMPLETION_ONSTACK(done);
1236 struct wb_writeback_work work = {
Wu Fengguang6e6938b2010-06-06 10:38:15 -06001237 .sb = sb,
1238 .sync_mode = WB_SYNC_NONE,
1239 .tagged_writepages = 1,
1240 .done = &done,
1241 .nr_pages = nr,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001242 .reason = reason,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001243 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001244
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001245 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001246 bdi_queue_work(sb->s_bdi, &work);
1247 wait_for_completion(&done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248}
Chris Mason3259f8b2010-10-29 11:16:17 -04001249EXPORT_SYMBOL(writeback_inodes_sb_nr);
1250
1251/**
1252 * writeback_inodes_sb - writeback dirty inodes from given super_block
1253 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001254 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001255 *
1256 * Start writeback on some inodes on this super_block. No guarantees are made
1257 * on how many (if any) will be written, and this function does not wait
1258 * for IO completion of submitted IO.
1259 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001260void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001261{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001262 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001263}
Jens Axboed8a85592009-09-02 12:34:32 +02001264EXPORT_SYMBOL(writeback_inodes_sb);
1265
1266/**
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001267 * writeback_inodes_sb_if_idle - start writeback if none underway
1268 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001269 * @reason: reason why some writeback work was initiated
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001270 *
1271 * Invoke writeback_inodes_sb if no writeback is currently underway.
1272 * Returns 1 if writeback was started, 0 if not.
1273 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001274int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason)
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001275{
1276 if (!writeback_in_progress(sb->s_bdi)) {
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001277 down_read(&sb->s_umount);
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001278 writeback_inodes_sb(sb, reason);
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001279 up_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001280 return 1;
1281 } else
1282 return 0;
1283}
1284EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1285
1286/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001287 * writeback_inodes_sb_if_idle - start writeback if none underway
1288 * @sb: the superblock
1289 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001290 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001291 *
1292 * Invoke writeback_inodes_sb if no writeback is currently underway.
1293 * Returns 1 if writeback was started, 0 if not.
1294 */
1295int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001296 unsigned long nr,
1297 enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001298{
1299 if (!writeback_in_progress(sb->s_bdi)) {
1300 down_read(&sb->s_umount);
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001301 writeback_inodes_sb_nr(sb, nr, reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001302 up_read(&sb->s_umount);
1303 return 1;
1304 } else
1305 return 0;
1306}
1307EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1308
1309/**
Jens Axboed8a85592009-09-02 12:34:32 +02001310 * sync_inodes_sb - sync sb inode pages
1311 * @sb: the superblock
1312 *
1313 * This function writes and waits on any dirty inode belonging to this
Stefan Hajnoczicb9ef8d2011-01-13 15:47:26 -08001314 * super_block.
Jens Axboed8a85592009-09-02 12:34:32 +02001315 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001316void sync_inodes_sb(struct super_block *sb)
Jens Axboed8a85592009-09-02 12:34:32 +02001317{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001318 DECLARE_COMPLETION_ONSTACK(done);
1319 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001320 .sb = sb,
1321 .sync_mode = WB_SYNC_ALL,
1322 .nr_pages = LONG_MAX,
1323 .range_cyclic = 0,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001324 .done = &done,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001325 .reason = WB_REASON_SYNC,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001326 };
1327
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001328 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1329
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001330 bdi_queue_work(sb->s_bdi, &work);
1331 wait_for_completion(&done);
1332
Jens Axboeb6e51312009-09-16 15:13:54 +02001333 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001334}
1335EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001338 * write_inode_now - write an inode to disk
1339 * @inode: inode to write to disk
1340 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001342 * This function commits an inode to disk immediately if it is dirty. This is
1343 * primarily needed by knfsd.
1344 *
1345 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347int write_inode_now(struct inode *inode, int sync)
1348{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001349 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 int ret;
1351 struct writeback_control wbc = {
1352 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001353 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001354 .range_start = 0,
1355 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 };
1357
1358 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001359 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
1361 might_sleep();
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001362 spin_lock(&wb->list_lock);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001363 spin_lock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001364 ret = writeback_single_inode(inode, wb, &wbc);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001365 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001366 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 if (sync)
Joern Engel1c0eeaf2007-10-16 23:30:44 -07001368 inode_sync_wait(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 return ret;
1370}
1371EXPORT_SYMBOL(write_inode_now);
1372
1373/**
1374 * sync_inode - write an inode and its pages to disk.
1375 * @inode: the inode to sync
1376 * @wbc: controls the writeback mode
1377 *
1378 * sync_inode() will write an inode and its pages to disk. It will also
1379 * correctly update the inode on its superblock's dirty inode lists and will
1380 * update inode->i_state.
1381 *
1382 * The caller must have a ref on the inode.
1383 */
1384int sync_inode(struct inode *inode, struct writeback_control *wbc)
1385{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001386 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 int ret;
1388
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001389 spin_lock(&wb->list_lock);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001390 spin_lock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001391 ret = writeback_single_inode(inode, wb, wbc);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001392 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001393 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 return ret;
1395}
1396EXPORT_SYMBOL(sync_inode);
Christoph Hellwigc3765012010-10-06 10:48:20 +02001397
1398/**
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001399 * sync_inode_metadata - write an inode to disk
Christoph Hellwigc3765012010-10-06 10:48:20 +02001400 * @inode: the inode to sync
1401 * @wait: wait for I/O to complete.
1402 *
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001403 * Write an inode to disk and adjust its dirty state after completion.
Christoph Hellwigc3765012010-10-06 10:48:20 +02001404 *
1405 * Note: only writes the actual inode, no associated data or other metadata.
1406 */
1407int sync_inode_metadata(struct inode *inode, int wait)
1408{
1409 struct writeback_control wbc = {
1410 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1411 .nr_to_write = 0, /* metadata-only */
1412 };
1413
1414 return sync_inode(inode, &wbc);
1415}
1416EXPORT_SYMBOL(sync_inode_metadata);