blob: 2627f0dfcd9c798243e31daa402e6d4bd37a7cf0 [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>
Jens Axboe03ba3782009-09-09 09:08:54 +020023#include <linux/kthread.h>
24#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/writeback.h>
26#include <linux/blkdev.h>
27#include <linux/backing-dev.h>
28#include <linux/buffer_head.h>
David Howells07f3f052006-09-30 20:52:18 +020029#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
Jens Axboe66f3b8e2009-09-02 09:19:46 +020031#define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info)
Adrian Bunkf11b00f2008-04-29 00:58:56 -070032
Jens Axboe03ba3782009-09-09 09:08:54 +020033/*
Jens Axboed0bceac2009-05-18 08:20:32 +020034 * We don't actually have pdflush, but this one is exported though /proc...
35 */
36int nr_pdflush_threads;
37
38/*
Jens Axboec4a77a62009-09-16 15:18:25 +020039 * Passed into wb_writeback(), essentially a subset of writeback_control
40 */
41struct wb_writeback_args {
42 long nr_pages;
43 struct super_block *sb;
44 enum writeback_sync_modes sync_mode;
H Hartley Sweeten52957fe2010-04-01 20:36:30 -050045 unsigned int for_kupdate:1;
46 unsigned int range_cyclic:1;
47 unsigned int for_background:1;
Jens Axboec4a77a62009-09-16 15:18:25 +020048};
49
50/*
Jens Axboe03ba3782009-09-09 09:08:54 +020051 * Work items for the bdi_writeback threads
Adrian Bunkf11b00f2008-04-29 00:58:56 -070052 */
Jens Axboe03ba3782009-09-09 09:08:54 +020053struct bdi_work {
Jens Axboe8010c3b2009-09-15 20:04:57 +020054 struct list_head list; /* pending work list */
55 struct rcu_head rcu_head; /* for RCU free/clear of work */
Jens Axboe03ba3782009-09-09 09:08:54 +020056
Jens Axboe8010c3b2009-09-15 20:04:57 +020057 unsigned long seen; /* threads that have seen this work */
58 atomic_t pending; /* number of threads still to do work */
Jens Axboe03ba3782009-09-09 09:08:54 +020059
Jens Axboe8010c3b2009-09-15 20:04:57 +020060 struct wb_writeback_args args; /* writeback arguments */
Jens Axboe03ba3782009-09-09 09:08:54 +020061
Jens Axboe8010c3b2009-09-15 20:04:57 +020062 unsigned long state; /* flag bits, see WS_* */
Jens Axboe03ba3782009-09-09 09:08:54 +020063};
64
65enum {
Christoph Hellwig7f0e7be2010-06-08 18:14:34 +020066 WS_INPROGRESS = 0,
67 WS_ONSTACK,
Jens Axboe03ba3782009-09-09 09:08:54 +020068};
69
Jens Axboe03ba3782009-09-09 09:08:54 +020070static inline void bdi_work_init(struct bdi_work *work,
Jens Axboeb6e51312009-09-16 15:13:54 +020071 struct wb_writeback_args *args)
Jens Axboe03ba3782009-09-09 09:08:54 +020072{
73 INIT_RCU_HEAD(&work->rcu_head);
Jens Axboeb6e51312009-09-16 15:13:54 +020074 work->args = *args;
Christoph Hellwig7f0e7be2010-06-08 18:14:34 +020075 __set_bit(WS_INPROGRESS, &work->state);
Jens Axboe03ba3782009-09-09 09:08:54 +020076}
77
Adrian Bunkf11b00f2008-04-29 00:58:56 -070078/**
79 * writeback_in_progress - determine whether there is writeback in progress
80 * @bdi: the device's backing_dev_info structure.
81 *
Jens Axboe03ba3782009-09-09 09:08:54 +020082 * Determine whether there is writeback waiting to be handled against a
83 * backing device.
Adrian Bunkf11b00f2008-04-29 00:58:56 -070084 */
85int writeback_in_progress(struct backing_dev_info *bdi)
86{
Jens Axboe03ba3782009-09-09 09:08:54 +020087 return !list_empty(&bdi->work_list);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070088}
89
Jens Axboe03ba3782009-09-09 09:08:54 +020090static void bdi_work_free(struct rcu_head *head)
Nick Piggin4195f732009-05-28 09:01:15 +020091{
Jens Axboe03ba3782009-09-09 09:08:54 +020092 struct bdi_work *work = container_of(head, struct bdi_work, rcu_head);
Nick Piggin4195f732009-05-28 09:01:15 +020093
Christoph Hellwig7f0e7be2010-06-08 18:14:34 +020094 clear_bit(WS_INPROGRESS, &work->state);
95 smp_mb__after_clear_bit();
96 wake_up_bit(&work->state, WS_INPROGRESS);
97
98 if (!test_bit(WS_ONSTACK, &work->state))
Jens Axboe03ba3782009-09-09 09:08:54 +020099 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200100}
101
102static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work)
103{
104 /*
105 * The caller has retrieved the work arguments from this work,
106 * drop our reference. If this is the last ref, delete and free it
107 */
108 if (atomic_dec_and_test(&work->pending)) {
109 struct backing_dev_info *bdi = wb->bdi;
110
111 spin_lock(&bdi->wb_lock);
112 list_del_rcu(&work->list);
113 spin_unlock(&bdi->wb_lock);
114
Christoph Hellwig7f0e7be2010-06-08 18:14:34 +0200115 call_rcu(&work->rcu_head, bdi_work_free);
Nick Piggin4195f732009-05-28 09:01:15 +0200116 }
117}
118
Jens Axboe03ba3782009-09-09 09:08:54 +0200119static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120{
Jens Axboebcddc3f2009-09-13 20:07:36 +0200121 work->seen = bdi->wb_mask;
122 BUG_ON(!work->seen);
123 atomic_set(&work->pending, bdi->wb_cnt);
124 BUG_ON(!bdi->wb_cnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125
Jens Axboebcddc3f2009-09-13 20:07:36 +0200126 /*
Nick Piggindeed62e2009-09-15 21:32:58 +0200127 * list_add_tail_rcu() contains the necessary barriers to
128 * make sure the above stores are seen before the item is
129 * noticed on the list
Jens Axboebcddc3f2009-09-13 20:07:36 +0200130 */
Jens Axboebcddc3f2009-09-13 20:07:36 +0200131 spin_lock(&bdi->wb_lock);
132 list_add_tail_rcu(&work->list, &bdi->work_list);
133 spin_unlock(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134
135 /*
Jens Axboe03ba3782009-09-09 09:08:54 +0200136 * If the default thread isn't there, make sure we add it. When
137 * it gets created and wakes up, we'll run this work.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200139 if (unlikely(list_empty_careful(&bdi->wb_list)))
140 wake_up_process(default_backing_dev_info.wb.task);
141 else {
142 struct bdi_writeback *wb = &bdi->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
Nick Piggin1ef7d9a2009-09-15 21:37:55 +0200144 if (wb->task)
Jens Axboe03ba3782009-09-09 09:08:54 +0200145 wake_up_process(wb->task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147}
148
Jens Axboe03ba3782009-09-09 09:08:54 +0200149/*
150 * Used for on-stack allocated work items. The caller needs to wait until
151 * the wb threads have acked the work before it's safe to continue.
152 */
Christoph Hellwig7f0e7be2010-06-08 18:14:34 +0200153static void bdi_wait_on_work_done(struct bdi_work *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154{
Christoph Hellwig7f0e7be2010-06-08 18:14:34 +0200155 wait_on_bit(&work->state, WS_INPROGRESS, bdi_sched_wait,
Jens Axboe03ba3782009-09-09 09:08:54 +0200156 TASK_UNINTERRUPTIBLE);
157}
158
Jens Axboef11fcae2009-09-15 09:53:35 +0200159static void bdi_alloc_queue_work(struct backing_dev_info *bdi,
Jens Axboef17625b2010-06-01 11:05:22 +0200160 struct wb_writeback_args *args)
Jens Axboe03ba3782009-09-09 09:08:54 +0200161{
162 struct bdi_work *work;
163
Jens Axboebcddc3f2009-09-13 20:07:36 +0200164 /*
165 * This is WB_SYNC_NONE writeback, so if allocation fails just
166 * wakeup the thread for old dirty data writeback
167 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200168 work = kmalloc(sizeof(*work), GFP_ATOMIC);
Jens Axboebcddc3f2009-09-13 20:07:36 +0200169 if (work) {
Jens Axboeb6e51312009-09-16 15:13:54 +0200170 bdi_work_init(work, args);
Jens Axboebcddc3f2009-09-13 20:07:36 +0200171 bdi_queue_work(bdi, work);
172 } else {
173 struct bdi_writeback *wb = &bdi->wb;
Jens Axboe03ba3782009-09-09 09:08:54 +0200174
Jens Axboebcddc3f2009-09-13 20:07:36 +0200175 if (wb->task)
176 wake_up_process(wb->task);
177 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200178}
179
Jens Axboeb6e51312009-09-16 15:13:54 +0200180/**
Christoph Hellwig3c4d7162010-06-08 18:14:43 +0200181 * bdi_queue_work_onstack - start and wait for writeback
Jens Axboeb6e51312009-09-16 15:13:54 +0200182 * @sb: write inodes from this super_block
183 *
184 * Description:
Christoph Hellwig3c4d7162010-06-08 18:14:43 +0200185 * This function initiates writeback and waits for the operation to
186 * complete. Callers must hold the sb s_umount semaphore for
Jens Axboeb6e51312009-09-16 15:13:54 +0200187 * reading, to avoid having the super disappear before we are done.
188 */
Christoph Hellwig3c4d7162010-06-08 18:14:43 +0200189static void bdi_queue_work_onstack(struct wb_writeback_args *args)
Jens Axboe03ba3782009-09-09 09:08:54 +0200190{
Jens Axboeb6e51312009-09-16 15:13:54 +0200191 struct bdi_work work;
Christoph Hellwigf0fad8a2009-09-11 09:47:56 +0200192
Christoph Hellwig3c4d7162010-06-08 18:14:43 +0200193 bdi_work_init(&work, args);
Christoph Hellwig7f0e7be2010-06-08 18:14:34 +0200194 __set_bit(WS_ONSTACK, &work.state);
Christoph Hellwigf0fad8a2009-09-11 09:47:56 +0200195
Christoph Hellwig3c4d7162010-06-08 18:14:43 +0200196 bdi_queue_work(args->sb->s_bdi, &work);
Christoph Hellwig7f0e7be2010-06-08 18:14:34 +0200197 bdi_wait_on_work_done(&work);
Jens Axboeb6e51312009-09-16 15:13:54 +0200198}
199
200/**
201 * bdi_start_writeback - start writeback
202 * @bdi: the backing device to write from
Jaswinder Singh Rajput4b6764f2010-01-01 20:35:23 -0800203 * @sb: write inodes from this super_block
Jens Axboeb6e51312009-09-16 15:13:54 +0200204 * @nr_pages: the number of pages to write
205 *
206 * Description:
207 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
208 * started when this function returns, we make no guarentees on
Jens Axboe0e3c9a22010-06-01 11:08:43 +0200209 * completion. Caller need not hold sb s_umount semaphore.
Jens Axboeb6e51312009-09-16 15:13:54 +0200210 *
211 */
Jens Axboea72bfd42009-09-26 00:07:46 +0200212void bdi_start_writeback(struct backing_dev_info *bdi, struct super_block *sb,
Jens Axboe0e3c9a22010-06-01 11:08:43 +0200213 long nr_pages)
Jens Axboeb6e51312009-09-16 15:13:54 +0200214{
215 struct wb_writeback_args args = {
Jens Axboea72bfd42009-09-26 00:07:46 +0200216 .sb = sb,
Jens Axboeb6e51312009-09-16 15:13:54 +0200217 .sync_mode = WB_SYNC_NONE,
218 .nr_pages = nr_pages,
219 .range_cyclic = 1,
220 };
221
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800222 /*
223 * We treat @nr_pages=0 as the special case to do background writeback,
224 * ie. to sync pages until the background dirty threshold is reached.
225 */
226 if (!nr_pages) {
227 args.nr_pages = LONG_MAX;
228 args.for_background = 1;
229 }
230
Jens Axboef17625b2010-06-01 11:05:22 +0200231 bdi_alloc_queue_work(bdi, &args);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232}
233
234/*
Andrew Morton6610a0b2007-10-16 23:30:32 -0700235 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
236 * furthest end of its superblock's dirty-inode list.
237 *
238 * Before stamping the inode's ->dirtied_when, we check to see whether it is
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200239 * already the most-recently-dirtied inode on the b_dirty list. If that is
Andrew Morton6610a0b2007-10-16 23:30:32 -0700240 * the case then the inode must have been redirtied while it was being written
241 * out and we don't reset its dirtied_when.
242 */
243static void redirty_tail(struct inode *inode)
244{
Jens Axboe03ba3782009-09-09 09:08:54 +0200245 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700246
Jens Axboe03ba3782009-09-09 09:08:54 +0200247 if (!list_empty(&wb->b_dirty)) {
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200248 struct inode *tail;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700249
Jens Axboe03ba3782009-09-09 09:08:54 +0200250 tail = list_entry(wb->b_dirty.next, struct inode, i_list);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200251 if (time_before(inode->dirtied_when, tail->dirtied_when))
Andrew Morton6610a0b2007-10-16 23:30:32 -0700252 inode->dirtied_when = jiffies;
253 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200254 list_move(&inode->i_list, &wb->b_dirty);
Andrew Morton6610a0b2007-10-16 23:30:32 -0700255}
256
257/*
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200258 * requeue inode for re-scanning after bdi->b_io list is exhausted.
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700259 */
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700260static void requeue_io(struct inode *inode)
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700261{
Jens Axboe03ba3782009-09-09 09:08:54 +0200262 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
263
264 list_move(&inode->i_list, &wb->b_more_io);
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700265}
266
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700267static void inode_sync_complete(struct inode *inode)
268{
269 /*
270 * Prevent speculative execution through spin_unlock(&inode_lock);
271 */
272 smp_mb();
273 wake_up_bit(&inode->i_state, __I_SYNC);
274}
275
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700276static bool inode_dirtied_after(struct inode *inode, unsigned long t)
277{
278 bool ret = time_after(inode->dirtied_when, t);
279#ifndef CONFIG_64BIT
280 /*
281 * For inodes being constantly redirtied, dirtied_when can get stuck.
282 * It _appears_ to be in the future, but is actually in distant past.
283 * This test is necessary to prevent such wrapped-around relative times
Jens Axboe5b0830c2009-09-23 19:37:09 +0200284 * from permanently stopping the whole bdi writeback.
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700285 */
286 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
287#endif
288 return ret;
289}
290
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700291/*
Fengguang Wu2c136572007-10-16 23:30:39 -0700292 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
293 */
294static void move_expired_inodes(struct list_head *delaying_queue,
295 struct list_head *dispatch_queue,
296 unsigned long *older_than_this)
297{
Shaohua Li5c034492009-09-24 14:42:33 +0200298 LIST_HEAD(tmp);
299 struct list_head *pos, *node;
Jens Axboecf137302009-09-24 15:12:57 +0200300 struct super_block *sb = NULL;
Shaohua Li5c034492009-09-24 14:42:33 +0200301 struct inode *inode;
Jens Axboecf137302009-09-24 15:12:57 +0200302 int do_sb_sort = 0;
Shaohua Li5c034492009-09-24 14:42:33 +0200303
Fengguang Wu2c136572007-10-16 23:30:39 -0700304 while (!list_empty(delaying_queue)) {
Shaohua Li5c034492009-09-24 14:42:33 +0200305 inode = list_entry(delaying_queue->prev, struct inode, i_list);
Fengguang Wu2c136572007-10-16 23:30:39 -0700306 if (older_than_this &&
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700307 inode_dirtied_after(inode, *older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700308 break;
Jens Axboecf137302009-09-24 15:12:57 +0200309 if (sb && sb != inode->i_sb)
310 do_sb_sort = 1;
311 sb = inode->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200312 list_move(&inode->i_list, &tmp);
313 }
314
Jens Axboecf137302009-09-24 15:12:57 +0200315 /* just one sb in list, splice to dispatch_queue and we're done */
316 if (!do_sb_sort) {
317 list_splice(&tmp, dispatch_queue);
318 return;
319 }
320
Shaohua Li5c034492009-09-24 14:42:33 +0200321 /* Move inodes from one superblock together */
322 while (!list_empty(&tmp)) {
323 inode = list_entry(tmp.prev, struct inode, i_list);
324 sb = inode->i_sb;
325 list_for_each_prev_safe(pos, node, &tmp) {
326 inode = list_entry(pos, struct inode, i_list);
327 if (inode->i_sb == sb)
328 list_move(&inode->i_list, dispatch_queue);
329 }
Fengguang Wu2c136572007-10-16 23:30:39 -0700330 }
331}
332
333/*
334 * Queue all expired dirty inodes for io, eldest first.
335 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200336static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
Fengguang Wu2c136572007-10-16 23:30:39 -0700337{
Jens Axboe03ba3782009-09-09 09:08:54 +0200338 list_splice_init(&wb->b_more_io, wb->b_io.prev);
339 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200340}
341
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100342static int write_inode(struct inode *inode, struct writeback_control *wbc)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200343{
Jens Axboe03ba3782009-09-09 09:08:54 +0200344 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100345 return inode->i_sb->s_op->write_inode(inode, wbc);
Jens Axboe03ba3782009-09-09 09:08:54 +0200346 return 0;
Fengguang Wu2c136572007-10-16 23:30:39 -0700347}
348
349/*
Christoph Hellwig01c03192009-06-08 13:35:40 +0200350 * Wait for writeback on an inode to complete.
351 */
352static void inode_wait_for_writeback(struct inode *inode)
353{
354 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
355 wait_queue_head_t *wqh;
356
357 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Richard Kennedy58a9d3d82010-05-24 14:32:38 -0700358 while (inode->i_state & I_SYNC) {
Christoph Hellwig01c03192009-06-08 13:35:40 +0200359 spin_unlock(&inode_lock);
360 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
361 spin_lock(&inode_lock);
Richard Kennedy58a9d3d82010-05-24 14:32:38 -0700362 }
Christoph Hellwig01c03192009-06-08 13:35:40 +0200363}
364
365/*
366 * Write out an inode's dirty pages. Called under inode_lock. Either the
367 * caller has ref on the inode (either via __iget or via syscall against an fd)
368 * or the inode has I_WILL_FREE set (via generic_forget_inode)
369 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 * If `wait' is set, wait on the writeout.
371 *
372 * The whole writeout design is quite complex and fragile. We want to avoid
373 * starvation of particular inodes when others are being redirtied, prevent
374 * livelocks, etc.
375 *
376 * Called under inode_lock.
377 */
378static int
Christoph Hellwig01c03192009-06-08 13:35:40 +0200379writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 struct address_space *mapping = inode->i_mapping;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200382 unsigned dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 int ret;
384
Christoph Hellwig01c03192009-06-08 13:35:40 +0200385 if (!atomic_read(&inode->i_count))
386 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
387 else
388 WARN_ON(inode->i_state & I_WILL_FREE);
389
390 if (inode->i_state & I_SYNC) {
391 /*
392 * If this inode is locked for writeback and we are not doing
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200393 * writeback-for-data-integrity, move it to b_more_io so that
Christoph Hellwig01c03192009-06-08 13:35:40 +0200394 * writeback can proceed with the other inodes on s_io.
395 *
396 * We'll have another go at writing back this inode when we
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200397 * completed a full scan of b_io.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200398 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100399 if (wbc->sync_mode != WB_SYNC_ALL) {
Christoph Hellwig01c03192009-06-08 13:35:40 +0200400 requeue_io(inode);
401 return 0;
402 }
403
404 /*
405 * It's a data-integrity sync. We must wait.
406 */
407 inode_wait_for_writeback(inode);
408 }
409
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700410 BUG_ON(inode->i_state & I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400412 /* Set I_SYNC, reset I_DIRTY_PAGES */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700413 inode->i_state |= I_SYNC;
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400414 inode->i_state &= ~I_DIRTY_PAGES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 spin_unlock(&inode_lock);
416
417 ret = do_writepages(mapping, wbc);
418
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100419 /*
420 * Make sure to wait on the data before writing out the metadata.
421 * This is important for filesystems that modify metadata on data
422 * I/O completion.
423 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100424 if (wbc->sync_mode == WB_SYNC_ALL) {
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100425 int err = filemap_fdatawait(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 if (ret == 0)
427 ret = err;
428 }
429
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400430 /*
431 * Some filesystems may redirty the inode during the writeback
432 * due to delalloc, clear dirty metadata flags right before
433 * write_inode()
434 */
435 spin_lock(&inode_lock);
436 dirty = inode->i_state & I_DIRTY;
437 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
438 spin_unlock(&inode_lock);
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100439 /* Don't write the inode if only I_DIRTY_PAGES was set */
440 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100441 int err = write_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 if (ret == 0)
443 ret = err;
444 }
445
446 spin_lock(&inode_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700447 inode->i_state &= ~I_SYNC;
Wu Fengguang84a89242009-06-16 15:33:17 -0700448 if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
Wu Fengguangb3af9462009-09-25 06:04:10 +0200449 if ((inode->i_state & I_DIRTY_PAGES) && wbc->for_kupdate) {
Wu Fengguangae1b7f72009-09-23 20:33:42 +0800450 /*
Wu Fengguangb3af9462009-09-25 06:04:10 +0200451 * More pages get dirtied by a fast dirtier.
452 */
453 goto select_queue;
454 } else if (inode->i_state & I_DIRTY) {
455 /*
456 * At least XFS will redirty the inode during the
457 * writeback (delalloc) and on io completion (isize).
Wu Fengguangae1b7f72009-09-23 20:33:42 +0800458 */
459 redirty_tail(inode);
460 } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 /*
462 * We didn't write back all the pages. nfs_writepages()
463 * sometimes bales out without doing anything. Redirty
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200464 * the inode; Move it from b_io onto b_more_io/b_dirty.
Andrew Morton1b43ef92007-10-16 23:30:35 -0700465 */
466 /*
467 * akpm: if the caller was the kupdate function we put
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200468 * this inode at the head of b_dirty so it gets first
Andrew Morton1b43ef92007-10-16 23:30:35 -0700469 * consideration. Otherwise, move it to the tail, for
470 * the reasons described there. I'm not really sure
471 * how much sense this makes. Presumably I had a good
472 * reasons for doing it this way, and I'd rather not
473 * muck with it at present.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 */
475 if (wbc->for_kupdate) {
476 /*
Fengguang Wu2c136572007-10-16 23:30:39 -0700477 * For the kupdate function we move the inode
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200478 * to b_more_io so it will get more writeout as
Fengguang Wu2c136572007-10-16 23:30:39 -0700479 * soon as the queue becomes uncongested.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 */
481 inode->i_state |= I_DIRTY_PAGES;
Wu Fengguangb3af9462009-09-25 06:04:10 +0200482select_queue:
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800483 if (wbc->nr_to_write <= 0) {
484 /*
485 * slice used up: queue for next turn
486 */
487 requeue_io(inode);
488 } else {
489 /*
490 * somehow blocked: retry later
491 */
492 redirty_tail(inode);
493 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 } else {
495 /*
496 * Otherwise fully redirty the inode so that
497 * other inodes on this superblock will get some
498 * writeout. Otherwise heavy writing to one
499 * file would indefinitely suspend writeout of
500 * all the other files.
501 */
502 inode->i_state |= I_DIRTY_PAGES;
Andrew Morton1b43ef92007-10-16 23:30:35 -0700503 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 } else if (atomic_read(&inode->i_count)) {
506 /*
507 * The inode is clean, inuse
508 */
509 list_move(&inode->i_list, &inode_in_use);
510 } else {
511 /*
512 * The inode is clean, unused
513 */
514 list_move(&inode->i_list, &inode_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 }
516 }
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700517 inode_sync_complete(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 return ret;
519}
520
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800521static void unpin_sb_for_writeback(struct super_block *sb)
Jens Axboe9ecc2732009-09-24 15:25:11 +0200522{
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800523 up_read(&sb->s_umount);
524 put_super(sb);
Jens Axboe9ecc2732009-09-24 15:25:11 +0200525}
526
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800527enum sb_pin_state {
528 SB_PINNED,
529 SB_NOT_PINNED,
530 SB_PIN_FAILED
531};
532
Jens Axboe03ba3782009-09-09 09:08:54 +0200533/*
534 * For WB_SYNC_NONE writeback, the caller does not have the sb pinned
535 * before calling writeback. So make sure that we do pin it, so it doesn't
536 * go away while we are writing inodes from it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200537 */
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800538static enum sb_pin_state pin_sb_for_writeback(struct writeback_control *wbc,
539 struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540{
Jens Axboe9ecc2732009-09-24 15:25:11 +0200541 /*
Jens Axboe03ba3782009-09-09 09:08:54 +0200542 * Caller must already hold the ref for this
543 */
Jens Axboe0e3c9a22010-06-01 11:08:43 +0200544 if (wbc->sync_mode == WB_SYNC_ALL) {
Jens Axboe03ba3782009-09-09 09:08:54 +0200545 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800546 return SB_NOT_PINNED;
Jens Axboe03ba3782009-09-09 09:08:54 +0200547 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200548 spin_lock(&sb_lock);
549 sb->s_count++;
550 if (down_read_trylock(&sb->s_umount)) {
551 if (sb->s_root) {
552 spin_unlock(&sb_lock);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800553 return SB_PINNED;
Jens Axboe03ba3782009-09-09 09:08:54 +0200554 }
555 /*
556 * umounted, drop rwsem again and fall through to failure
557 */
558 up_read(&sb->s_umount);
559 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200560 sb->s_count--;
561 spin_unlock(&sb_lock);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800562 return SB_PIN_FAILED;
Jens Axboe03ba3782009-09-09 09:08:54 +0200563}
564
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800565/*
566 * Write a portion of b_io inodes which belong to @sb.
567 * If @wbc->sb != NULL, then find and write all such
568 * inodes. Otherwise write only ones which go sequentially
569 * in reverse order.
570 * Return 1, if the caller writeback routine should be
571 * interrupted. Otherwise return 0.
572 */
573static int writeback_sb_inodes(struct super_block *sb,
574 struct bdi_writeback *wb,
575 struct writeback_control *wbc)
Jens Axboe03ba3782009-09-09 09:08:54 +0200576{
Jens Axboe03ba3782009-09-09 09:08:54 +0200577 while (!list_empty(&wb->b_io)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 long pages_skipped;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800579 struct inode *inode = list_entry(wb->b_io.prev,
580 struct inode, i_list);
581 if (wbc->sb && sb != inode->i_sb) {
582 /* super block given and doesn't
583 match, skip this inode */
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200584 redirty_tail(inode);
585 continue;
586 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800587 if (sb != inode->i_sb)
588 /* finish with this superblock */
589 return 0;
Wu Fengguang84a89242009-06-16 15:33:17 -0700590 if (inode->i_state & (I_NEW | I_WILL_FREE)) {
Nick Piggin7ef0d732009-03-12 14:31:38 -0700591 requeue_io(inode);
592 continue;
593 }
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700594 /*
595 * Was this inode dirtied after sync_sb_inodes was called?
596 * This keeps sync from extra jobs and livelock.
597 */
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800598 if (inode_dirtied_after(inode, wbc->wb_start))
599 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
Wu Fengguang84a89242009-06-16 15:33:17 -0700601 BUG_ON(inode->i_state & (I_FREEING | I_CLEAR));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 __iget(inode);
603 pages_skipped = wbc->pages_skipped;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200604 writeback_single_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 if (wbc->pages_skipped != pages_skipped) {
606 /*
607 * writeback is not making progress due to locked
608 * buffers. Skip this inode for now.
609 */
Andrew Mortonf57b9b72007-10-16 23:30:34 -0700610 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 }
612 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 iput(inode);
OGAWA Hirofumi4ffc8442006-03-25 03:07:44 -0800614 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 spin_lock(&inode_lock);
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800616 if (wbc->nr_to_write <= 0) {
617 wbc->more_io = 1;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800618 return 1;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800619 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200620 if (!list_empty(&wb->b_more_io))
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800621 wbc->more_io = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800623 /* b_io is empty */
624 return 1;
625}
Nick Piggin38f21972009-01-06 14:40:25 -0800626
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800627static void writeback_inodes_wb(struct bdi_writeback *wb,
628 struct writeback_control *wbc)
629{
630 int ret = 0;
Jens Axboe9ecc2732009-09-24 15:25:11 +0200631
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800632 wbc->wb_start = jiffies; /* livelock avoidance */
633 spin_lock(&inode_lock);
634 if (!wbc->for_kupdate || list_empty(&wb->b_io))
635 queue_io(wb, wbc->older_than_this);
636
637 while (!list_empty(&wb->b_io)) {
638 struct inode *inode = list_entry(wb->b_io.prev,
639 struct inode, i_list);
640 struct super_block *sb = inode->i_sb;
641 enum sb_pin_state state;
642
643 if (wbc->sb && sb != wbc->sb) {
644 /* super block given and doesn't
645 match, skip this inode */
646 redirty_tail(inode);
647 continue;
648 }
649 state = pin_sb_for_writeback(wbc, sb);
650
651 if (state == SB_PIN_FAILED) {
652 requeue_io(inode);
653 continue;
654 }
655 ret = writeback_sb_inodes(sb, wb, wbc);
656
657 if (state == SB_PINNED)
658 unpin_sb_for_writeback(sb);
659 if (ret)
660 break;
661 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200662 spin_unlock(&inode_lock);
663 /* Leave any unwritten inodes on b_io */
664}
665
Jens Axboe03ba3782009-09-09 09:08:54 +0200666void writeback_inodes_wbc(struct writeback_control *wbc)
667{
668 struct backing_dev_info *bdi = wbc->bdi;
669
670 writeback_inodes_wb(&bdi->wb, wbc);
671}
672
673/*
674 * The maximum number of pages to writeout in a single bdi flush/kupdate
675 * operation. We do this so we don't hold I_SYNC against an inode for
676 * enormous amounts of time, which would block a userspace task which has
677 * been forced to throttle against that inode. Also, the code reevaluates
678 * the dirty each time it has written this many pages.
679 */
680#define MAX_WRITEBACK_PAGES 1024
681
682static inline bool over_bground_thresh(void)
683{
684 unsigned long background_thresh, dirty_thresh;
685
686 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL);
687
688 return (global_page_state(NR_FILE_DIRTY) +
689 global_page_state(NR_UNSTABLE_NFS) >= background_thresh);
690}
691
692/*
693 * Explicit flushing or periodic writeback of "old" data.
694 *
695 * Define "old": the first time one of an inode's pages is dirtied, we mark the
696 * dirtying-time in the inode's address_space. So this periodic writeback code
697 * just walks the superblock inode list, writing back any inodes which are
698 * older than a specific point in time.
699 *
700 * Try to run once per dirty_writeback_interval. But if a writeback event
701 * takes longer than a dirty_writeback_interval interval, then leave a
702 * one-second gap.
703 *
704 * older_than_this takes precedence over nr_to_write. So we'll only write back
705 * all dirty pages if they are all attached to "old" mappings.
706 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200707static long wb_writeback(struct bdi_writeback *wb,
708 struct wb_writeback_args *args)
Jens Axboe03ba3782009-09-09 09:08:54 +0200709{
710 struct writeback_control wbc = {
711 .bdi = wb->bdi,
Jens Axboec4a77a62009-09-16 15:18:25 +0200712 .sb = args->sb,
713 .sync_mode = args->sync_mode,
Jens Axboe03ba3782009-09-09 09:08:54 +0200714 .older_than_this = NULL,
Jens Axboec4a77a62009-09-16 15:18:25 +0200715 .for_kupdate = args->for_kupdate,
Wu Fengguangb17621f2009-12-03 13:54:25 +0100716 .for_background = args->for_background,
Jens Axboec4a77a62009-09-16 15:18:25 +0200717 .range_cyclic = args->range_cyclic,
Jens Axboe03ba3782009-09-09 09:08:54 +0200718 };
719 unsigned long oldest_jif;
720 long wrote = 0;
Jan Karaa5989bd2009-09-16 19:22:48 +0200721 struct inode *inode;
Jens Axboe03ba3782009-09-09 09:08:54 +0200722
723 if (wbc.for_kupdate) {
724 wbc.older_than_this = &oldest_jif;
725 oldest_jif = jiffies -
726 msecs_to_jiffies(dirty_expire_interval * 10);
727 }
Jens Axboec4a77a62009-09-16 15:18:25 +0200728 if (!wbc.range_cyclic) {
729 wbc.range_start = 0;
730 wbc.range_end = LLONG_MAX;
731 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200732
733 for (;;) {
734 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800735 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200736 */
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800737 if (args->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200738 break;
739
740 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800741 * For background writeout, stop when we are below the
742 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200743 */
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800744 if (args->for_background && !over_bground_thresh())
Jens Axboe03ba3782009-09-09 09:08:54 +0200745 break;
746
747 wbc.more_io = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200748 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
749 wbc.pages_skipped = 0;
750 writeback_inodes_wb(wb, &wbc);
Jens Axboec4a77a62009-09-16 15:18:25 +0200751 args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
Jens Axboe03ba3782009-09-09 09:08:54 +0200752 wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write;
753
754 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200755 * If we consumed everything, see if we have more
Jens Axboe03ba3782009-09-09 09:08:54 +0200756 */
Jens Axboe71fd05a2009-09-23 19:32:26 +0200757 if (wbc.nr_to_write <= 0)
758 continue;
759 /*
760 * Didn't write everything and we don't have more IO, bail
761 */
762 if (!wbc.more_io)
Jens Axboe03ba3782009-09-09 09:08:54 +0200763 break;
Jens Axboe71fd05a2009-09-23 19:32:26 +0200764 /*
765 * Did we write something? Try for more
766 */
767 if (wbc.nr_to_write < MAX_WRITEBACK_PAGES)
768 continue;
769 /*
770 * Nothing written. Wait for some inode to
771 * become available for writeback. Otherwise
772 * we'll just busyloop.
773 */
774 spin_lock(&inode_lock);
775 if (!list_empty(&wb->b_more_io)) {
776 inode = list_entry(wb->b_more_io.prev,
777 struct inode, i_list);
778 inode_wait_for_writeback(inode);
Jens Axboe03ba3782009-09-09 09:08:54 +0200779 }
Jens Axboe71fd05a2009-09-23 19:32:26 +0200780 spin_unlock(&inode_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200781 }
782
783 return wrote;
784}
785
786/*
787 * Return the next bdi_work struct that hasn't been processed by this
Jens Axboe8010c3b2009-09-15 20:04:57 +0200788 * wb thread yet. ->seen is initially set for each thread that exists
789 * for this device, when a thread first notices a piece of work it
790 * clears its bit. Depending on writeback type, the thread will notify
791 * completion on either receiving the work (WB_SYNC_NONE) or after
792 * it is done (WB_SYNC_ALL).
Jens Axboe03ba3782009-09-09 09:08:54 +0200793 */
794static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi,
795 struct bdi_writeback *wb)
796{
797 struct bdi_work *work, *ret = NULL;
798
799 rcu_read_lock();
800
801 list_for_each_entry_rcu(work, &bdi->work_list, list) {
Nick Piggin77fad5e2009-09-15 21:34:12 +0200802 if (!test_bit(wb->nr, &work->seen))
Jens Axboe03ba3782009-09-09 09:08:54 +0200803 continue;
Nick Piggin77fad5e2009-09-15 21:34:12 +0200804 clear_bit(wb->nr, &work->seen);
Jens Axboe03ba3782009-09-09 09:08:54 +0200805
806 ret = work;
807 break;
808 }
809
810 rcu_read_unlock();
811 return ret;
812}
813
814static long wb_check_old_data_flush(struct bdi_writeback *wb)
815{
816 unsigned long expired;
817 long nr_pages;
818
Jens Axboe69b62d02010-05-17 12:51:03 +0200819 /*
820 * When set to zero, disable periodic writeback
821 */
822 if (!dirty_writeback_interval)
823 return 0;
824
Jens Axboe03ba3782009-09-09 09:08:54 +0200825 expired = wb->last_old_flush +
826 msecs_to_jiffies(dirty_writeback_interval * 10);
827 if (time_before(jiffies, expired))
828 return 0;
829
830 wb->last_old_flush = jiffies;
831 nr_pages = global_page_state(NR_FILE_DIRTY) +
832 global_page_state(NR_UNSTABLE_NFS) +
833 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
834
Jens Axboec4a77a62009-09-16 15:18:25 +0200835 if (nr_pages) {
836 struct wb_writeback_args args = {
837 .nr_pages = nr_pages,
838 .sync_mode = WB_SYNC_NONE,
839 .for_kupdate = 1,
840 .range_cyclic = 1,
841 };
842
843 return wb_writeback(wb, &args);
844 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200845
846 return 0;
847}
848
849/*
850 * Retrieve work items and do the writeback they describe
851 */
852long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
853{
854 struct backing_dev_info *bdi = wb->bdi;
855 struct bdi_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +0200856 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200857
858 while ((work = get_next_work_item(bdi, wb)) != NULL) {
Jens Axboec4a77a62009-09-16 15:18:25 +0200859 struct wb_writeback_args args = work->args;
Jens Axboe03ba3782009-09-09 09:08:54 +0200860
861 /*
862 * Override sync mode, in case we must wait for completion
863 */
864 if (force_wait)
Jens Axboec4a77a62009-09-16 15:18:25 +0200865 work->args.sync_mode = args.sync_mode = WB_SYNC_ALL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200866
867 /*
868 * If this isn't a data integrity operation, just notify
869 * that we have seen this work and we are now starting it.
870 */
Christoph Hellwig7f0e7be2010-06-08 18:14:34 +0200871 if (!test_bit(WS_ONSTACK, &work->state))
Jens Axboe03ba3782009-09-09 09:08:54 +0200872 wb_clear_pending(wb, work);
873
Jens Axboec4a77a62009-09-16 15:18:25 +0200874 wrote += wb_writeback(wb, &args);
Jens Axboe03ba3782009-09-09 09:08:54 +0200875
876 /*
877 * This is a data integrity writeback, so only do the
878 * notification when we have completed the work.
879 */
Christoph Hellwig7f0e7be2010-06-08 18:14:34 +0200880 if (test_bit(WS_ONSTACK, &work->state))
Jens Axboe03ba3782009-09-09 09:08:54 +0200881 wb_clear_pending(wb, work);
882 }
883
884 /*
885 * Check for periodic writeback, kupdated() style
886 */
887 wrote += wb_check_old_data_flush(wb);
888
889 return wrote;
890}
891
892/*
893 * Handle writeback of dirty data for the device backed by this bdi. Also
894 * wakes up periodically and does kupdated style flushing.
895 */
896int bdi_writeback_task(struct bdi_writeback *wb)
897{
898 unsigned long last_active = jiffies;
899 unsigned long wait_jiffies = -1UL;
900 long pages_written;
901
902 while (!kthread_should_stop()) {
903 pages_written = wb_do_writeback(wb, 0);
904
905 if (pages_written)
906 last_active = jiffies;
907 else if (wait_jiffies != -1UL) {
908 unsigned long max_idle;
909
910 /*
911 * Longest period of inactivity that we tolerate. If we
912 * see dirty data again later, the task will get
913 * recreated automatically.
914 */
915 max_idle = max(5UL * 60 * HZ, wait_jiffies);
916 if (time_after(jiffies, max_idle + last_active))
917 break;
918 }
919
Jens Axboe69b62d02010-05-17 12:51:03 +0200920 if (dirty_writeback_interval) {
921 wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10);
922 schedule_timeout_interruptible(wait_jiffies);
Jens Axboef9eadbb2010-05-18 14:31:45 +0200923 } else {
924 set_current_state(TASK_INTERRUPTIBLE);
925 if (list_empty_careful(&wb->bdi->work_list) &&
926 !kthread_should_stop())
927 schedule();
928 __set_current_state(TASK_RUNNING);
929 }
Jens Axboe69b62d02010-05-17 12:51:03 +0200930
Jens Axboe03ba3782009-09-09 09:08:54 +0200931 try_to_freeze();
932 }
933
934 return 0;
935}
936
937/*
Jens Axboeb6e51312009-09-16 15:13:54 +0200938 * Schedule writeback for all backing devices. This does WB_SYNC_NONE
Christoph Hellwig3c4d7162010-06-08 18:14:43 +0200939 * writeback, for integrity writeback see bdi_queue_work_onstack().
Jens Axboe03ba3782009-09-09 09:08:54 +0200940 */
Jens Axboeb6e51312009-09-16 15:13:54 +0200941static void bdi_writeback_all(struct super_block *sb, long nr_pages)
Jens Axboe03ba3782009-09-09 09:08:54 +0200942{
Jens Axboeb6e51312009-09-16 15:13:54 +0200943 struct wb_writeback_args args = {
944 .sb = sb,
945 .nr_pages = nr_pages,
946 .sync_mode = WB_SYNC_NONE,
947 };
Jens Axboe03ba3782009-09-09 09:08:54 +0200948 struct backing_dev_info *bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +0200949
Jens Axboecfc4ba52009-09-14 13:12:40 +0200950 rcu_read_lock();
Jens Axboe03ba3782009-09-09 09:08:54 +0200951
Jens Axboecfc4ba52009-09-14 13:12:40 +0200952 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
Jens Axboe03ba3782009-09-09 09:08:54 +0200953 if (!bdi_has_dirty_io(bdi))
954 continue;
955
Jens Axboef17625b2010-06-01 11:05:22 +0200956 bdi_alloc_queue_work(bdi, &args);
Jens Axboe03ba3782009-09-09 09:08:54 +0200957 }
958
Jens Axboecfc4ba52009-09-14 13:12:40 +0200959 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +0200960}
961
962/*
963 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
964 * the whole world.
965 */
966void wakeup_flusher_threads(long nr_pages)
967{
Jens Axboe03ba3782009-09-09 09:08:54 +0200968 if (nr_pages == 0)
969 nr_pages = global_page_state(NR_FILE_DIRTY) +
970 global_page_state(NR_UNSTABLE_NFS);
Jens Axboeb6e51312009-09-16 15:13:54 +0200971 bdi_writeback_all(NULL, nr_pages);
Jens Axboe03ba3782009-09-09 09:08:54 +0200972}
973
974static noinline void block_dump___mark_inode_dirty(struct inode *inode)
975{
976 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
977 struct dentry *dentry;
978 const char *name = "?";
979
980 dentry = d_find_alias(inode);
981 if (dentry) {
982 spin_lock(&dentry->d_lock);
983 name = (const char *) dentry->d_name.name;
984 }
985 printk(KERN_DEBUG
986 "%s(%d): dirtied inode %lu (%s) on %s\n",
987 current->comm, task_pid_nr(current), inode->i_ino,
988 name, inode->i_sb->s_id);
989 if (dentry) {
990 spin_unlock(&dentry->d_lock);
991 dput(dentry);
992 }
993 }
994}
995
996/**
997 * __mark_inode_dirty - internal function
998 * @inode: inode to mark
999 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1000 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1001 * mark_inode_dirty_sync.
1002 *
1003 * Put the inode on the super block's dirty list.
1004 *
1005 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1006 * dirty list only if it is hashed or if it refers to a blockdev.
1007 * If it was not hashed, it will never be added to the dirty list
1008 * even if it is later hashed, as it will have been marked dirty already.
1009 *
1010 * In short, make sure you hash any inodes _before_ you start marking
1011 * them dirty.
1012 *
1013 * This function *must* be atomic for the I_DIRTY_PAGES case -
1014 * set_page_dirty() is called under spinlock in several places.
1015 *
1016 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1017 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1018 * the kernel-internal blockdev inode represents the dirtying time of the
1019 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1020 * page->mapping->host, so the page-dirtying time is recorded in the internal
1021 * blockdev inode.
1022 */
1023void __mark_inode_dirty(struct inode *inode, int flags)
1024{
1025 struct super_block *sb = inode->i_sb;
1026
1027 /*
1028 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1029 * dirty the inode itself
1030 */
1031 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1032 if (sb->s_op->dirty_inode)
1033 sb->s_op->dirty_inode(inode);
1034 }
1035
1036 /*
1037 * make sure that changes are seen by all cpus before we test i_state
1038 * -- mikulas
1039 */
1040 smp_mb();
1041
1042 /* avoid the locking if we can */
1043 if ((inode->i_state & flags) == flags)
1044 return;
1045
1046 if (unlikely(block_dump))
1047 block_dump___mark_inode_dirty(inode);
1048
1049 spin_lock(&inode_lock);
1050 if ((inode->i_state & flags) != flags) {
1051 const int was_dirty = inode->i_state & I_DIRTY;
1052
1053 inode->i_state |= flags;
1054
1055 /*
1056 * If the inode is being synced, just update its dirty state.
1057 * The unlocker will place the inode on the appropriate
1058 * superblock list, based upon its state.
1059 */
1060 if (inode->i_state & I_SYNC)
1061 goto out;
1062
1063 /*
1064 * Only add valid (hashed) inodes to the superblock's
1065 * dirty list. Add blockdev inodes as well.
1066 */
1067 if (!S_ISBLK(inode->i_mode)) {
1068 if (hlist_unhashed(&inode->i_hash))
1069 goto out;
1070 }
1071 if (inode->i_state & (I_FREEING|I_CLEAR))
1072 goto out;
1073
1074 /*
1075 * If the inode was already on b_dirty/b_io/b_more_io, don't
1076 * reposition it (that would break b_dirty time-ordering).
1077 */
1078 if (!was_dirty) {
1079 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Jens Axboe500b0672009-09-09 09:10:25 +02001080 struct backing_dev_info *bdi = wb->bdi;
1081
1082 if (bdi_cap_writeback_dirty(bdi) &&
1083 !test_bit(BDI_registered, &bdi->state)) {
1084 WARN_ON(1);
1085 printk(KERN_ERR "bdi-%s not registered\n",
1086 bdi->name);
1087 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001088
1089 inode->dirtied_when = jiffies;
1090 list_move(&inode->i_list, &wb->b_dirty);
1091 }
1092 }
1093out:
1094 spin_unlock(&inode_lock);
1095}
1096EXPORT_SYMBOL(__mark_inode_dirty);
1097
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001098/*
1099 * Write out a superblock's list of dirty inodes. A wait will be performed
1100 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1101 *
1102 * If older_than_this is non-NULL, then only write out inodes which
1103 * had their first dirtying at a time earlier than *older_than_this.
1104 *
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001105 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1106 * This function assumes that the blockdev superblock's inodes are backed by
1107 * a variety of queues, so all inodes are searched. For other superblocks,
1108 * assume that all inodes are backed by the same queue.
1109 *
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001110 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1111 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1112 * on the writer throttling path, and we get decent balancing between many
1113 * throttled threads: we don't want them all piling up on inode_sync_wait.
1114 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001115static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001116{
Jens Axboe03ba3782009-09-09 09:08:54 +02001117 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001118
Jens Axboe03ba3782009-09-09 09:08:54 +02001119 /*
1120 * We need to be protected against the filesystem going from
1121 * r/o to r/w or vice versa.
1122 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001123 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001124
Jens Axboe03ba3782009-09-09 09:08:54 +02001125 spin_lock(&inode_lock);
1126
1127 /*
1128 * Data integrity sync. Must wait for all pages under writeback,
1129 * because there may have been pages dirtied before our sync
1130 * call, but which had writeout started before we write it out.
1131 * In which case, the inode may not be on the dirty list, but
1132 * we still have to wait for that writeout.
1133 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001134 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Jens Axboe03ba3782009-09-09 09:08:54 +02001135 struct address_space *mapping;
1136
1137 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW))
1138 continue;
1139 mapping = inode->i_mapping;
1140 if (mapping->nrpages == 0)
1141 continue;
1142 __iget(inode);
1143 spin_unlock(&inode_lock);
1144 /*
1145 * We hold a reference to 'inode' so it couldn't have
1146 * been removed from s_inodes list while we dropped the
1147 * inode_lock. We cannot iput the inode now as we can
1148 * be holding the last reference and we cannot iput it
1149 * under inode_lock. So we keep the reference and iput
1150 * it later.
1151 */
1152 iput(old_inode);
1153 old_inode = inode;
1154
1155 filemap_fdatawait(mapping);
1156
1157 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001158
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001159 spin_lock(&inode_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001160 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001161 spin_unlock(&inode_lock);
1162 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163}
1164
Jens Axboed8a85592009-09-02 12:34:32 +02001165/**
1166 * writeback_inodes_sb - writeback dirty inodes from given super_block
1167 * @sb: the superblock
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 *
Jens Axboed8a85592009-09-02 12:34:32 +02001169 * Start writeback on some inodes on this super_block. No guarantees are made
1170 * on how many (if any) will be written, and this function does not wait
1171 * for IO completion of submitted IO. The number of pages submitted is
1172 * returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001174void writeback_inodes_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175{
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001176 unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY);
1177 unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS);
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001178 struct wb_writeback_args args = {
1179 .sb = sb,
1180 .sync_mode = WB_SYNC_NONE,
1181 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001182
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001183 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1184
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001185 args.nr_pages = nr_dirty + nr_unstable +
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001186 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
1187
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001188 bdi_queue_work_onstack(&args);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189}
Jens Axboed8a85592009-09-02 12:34:32 +02001190EXPORT_SYMBOL(writeback_inodes_sb);
1191
1192/**
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001193 * writeback_inodes_sb_if_idle - start writeback if none underway
1194 * @sb: the superblock
1195 *
1196 * Invoke writeback_inodes_sb if no writeback is currently underway.
1197 * Returns 1 if writeback was started, 0 if not.
1198 */
1199int writeback_inodes_sb_if_idle(struct super_block *sb)
1200{
1201 if (!writeback_in_progress(sb->s_bdi)) {
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001202 down_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001203 writeback_inodes_sb(sb);
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001204 up_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001205 return 1;
1206 } else
1207 return 0;
1208}
1209EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1210
1211/**
Jens Axboed8a85592009-09-02 12:34:32 +02001212 * sync_inodes_sb - sync sb inode pages
1213 * @sb: the superblock
1214 *
1215 * This function writes and waits on any dirty inode belonging to this
1216 * super_block. The number of pages synced is returned.
1217 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001218void sync_inodes_sb(struct super_block *sb)
Jens Axboed8a85592009-09-02 12:34:32 +02001219{
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001220 struct wb_writeback_args args = {
1221 .sb = sb,
1222 .sync_mode = WB_SYNC_ALL,
1223 .nr_pages = LONG_MAX,
1224 .range_cyclic = 0,
1225 };
1226
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001227 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1228
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001229 bdi_queue_work_onstack(&args);
Jens Axboeb6e51312009-09-16 15:13:54 +02001230 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001231}
1232EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001235 * write_inode_now - write an inode to disk
1236 * @inode: inode to write to disk
1237 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001239 * This function commits an inode to disk immediately if it is dirty. This is
1240 * primarily needed by knfsd.
1241 *
1242 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244int write_inode_now(struct inode *inode, int sync)
1245{
1246 int ret;
1247 struct writeback_control wbc = {
1248 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001249 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001250 .range_start = 0,
1251 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 };
1253
1254 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001255 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256
1257 might_sleep();
1258 spin_lock(&inode_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +02001259 ret = writeback_single_inode(inode, &wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 spin_unlock(&inode_lock);
1261 if (sync)
Joern Engel1c0eeaf2007-10-16 23:30:44 -07001262 inode_sync_wait(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 return ret;
1264}
1265EXPORT_SYMBOL(write_inode_now);
1266
1267/**
1268 * sync_inode - write an inode and its pages to disk.
1269 * @inode: the inode to sync
1270 * @wbc: controls the writeback mode
1271 *
1272 * sync_inode() will write an inode and its pages to disk. It will also
1273 * correctly update the inode on its superblock's dirty inode lists and will
1274 * update inode->i_state.
1275 *
1276 * The caller must have a ref on the inode.
1277 */
1278int sync_inode(struct inode *inode, struct writeback_control *wbc)
1279{
1280 int ret;
1281
1282 spin_lock(&inode_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +02001283 ret = writeback_single_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 spin_unlock(&inode_lock);
1285 return ret;
1286}
1287EXPORT_SYMBOL(sync_inode);