blob: 91013ff7dd5319dfd5c200461c3cdc617bb114e5 [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>
19#include <linux/sched.h>
20#include <linux/fs.h>
21#include <linux/mm.h>
22#include <linux/writeback.h>
23#include <linux/blkdev.h>
24#include <linux/backing-dev.h>
25#include <linux/buffer_head.h>
David Howells07f3f052006-09-30 20:52:18 +020026#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
Adrian Bunkf11b00f2008-04-29 00:58:56 -070028
29/**
30 * writeback_acquire - attempt to get exclusive writeback access to a device
31 * @bdi: the device's backing_dev_info structure
32 *
33 * It is a waste of resources to have more than one pdflush thread blocked on
34 * a single request queue. Exclusion at the request_queue level is obtained
35 * via a flag in the request_queue's backing_dev_info.state.
36 *
37 * Non-request_queue-backed address_spaces will share default_backing_dev_info,
38 * unless they implement their own. Which is somewhat inefficient, as this
39 * may prevent concurrent writeback against multiple devices.
40 */
41static int writeback_acquire(struct backing_dev_info *bdi)
42{
43 return !test_and_set_bit(BDI_pdflush, &bdi->state);
44}
45
46/**
47 * writeback_in_progress - determine whether there is writeback in progress
48 * @bdi: the device's backing_dev_info structure.
49 *
50 * Determine whether there is writeback in progress against a backing device.
51 */
52int writeback_in_progress(struct backing_dev_info *bdi)
53{
54 return test_bit(BDI_pdflush, &bdi->state);
55}
56
57/**
58 * writeback_release - relinquish exclusive writeback access against a device.
59 * @bdi: the device's backing_dev_info structure
60 */
61static void writeback_release(struct backing_dev_info *bdi)
62{
63 BUG_ON(!writeback_in_progress(bdi));
64 clear_bit(BDI_pdflush, &bdi->state);
65}
66
Linus Torvalds1da177e2005-04-16 15:20:36 -070067/**
68 * __mark_inode_dirty - internal function
69 * @inode: inode to mark
70 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
71 * Mark an inode as dirty. Callers should use mark_inode_dirty or
72 * mark_inode_dirty_sync.
73 *
74 * Put the inode on the super block's dirty list.
75 *
76 * CAREFUL! We mark it dirty unconditionally, but move it onto the
77 * dirty list only if it is hashed or if it refers to a blockdev.
78 * If it was not hashed, it will never be added to the dirty list
79 * even if it is later hashed, as it will have been marked dirty already.
80 *
81 * In short, make sure you hash any inodes _before_ you start marking
82 * them dirty.
83 *
84 * This function *must* be atomic for the I_DIRTY_PAGES case -
85 * set_page_dirty() is called under spinlock in several places.
86 *
87 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
88 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
89 * the kernel-internal blockdev inode represents the dirtying time of the
90 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
91 * page->mapping->host, so the page-dirtying time is recorded in the internal
92 * blockdev inode.
93 */
94void __mark_inode_dirty(struct inode *inode, int flags)
95{
96 struct super_block *sb = inode->i_sb;
97
98 /*
99 * Don't do this for I_DIRTY_PAGES - that doesn't actually
100 * dirty the inode itself
101 */
102 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
103 if (sb->s_op->dirty_inode)
104 sb->s_op->dirty_inode(inode);
105 }
106
107 /*
108 * make sure that changes are seen by all cpus before we test i_state
109 * -- mikulas
110 */
111 smp_mb();
112
113 /* avoid the locking if we can */
114 if ((inode->i_state & flags) == flags)
115 return;
116
117 if (unlikely(block_dump)) {
118 struct dentry *dentry = NULL;
119 const char *name = "?";
120
121 if (!list_empty(&inode->i_dentry)) {
122 dentry = list_entry(inode->i_dentry.next,
123 struct dentry, d_alias);
124 if (dentry && dentry->d_name.name)
125 name = (const char *) dentry->d_name.name;
126 }
127
128 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev"))
129 printk(KERN_DEBUG
130 "%s(%d): dirtied inode %lu (%s) on %s\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700131 current->comm, task_pid_nr(current), inode->i_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 name, inode->i_sb->s_id);
133 }
134
135 spin_lock(&inode_lock);
136 if ((inode->i_state & flags) != flags) {
137 const int was_dirty = inode->i_state & I_DIRTY;
138
139 inode->i_state |= flags;
140
141 /*
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700142 * If the inode is being synced, just update its dirty state.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 * The unlocker will place the inode on the appropriate
144 * superblock list, based upon its state.
145 */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700146 if (inode->i_state & I_SYNC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 goto out;
148
149 /*
150 * Only add valid (hashed) inodes to the superblock's
151 * dirty list. Add blockdev inodes as well.
152 */
153 if (!S_ISBLK(inode->i_mode)) {
154 if (hlist_unhashed(&inode->i_hash))
155 goto out;
156 }
157 if (inode->i_state & (I_FREEING|I_CLEAR))
158 goto out;
159
160 /*
Fengguang Wu2c136572007-10-16 23:30:39 -0700161 * If the inode was already on s_dirty/s_io/s_more_io, don't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 * reposition it (that would break s_dirty time-ordering).
163 */
164 if (!was_dirty) {
165 inode->dirtied_when = jiffies;
166 list_move(&inode->i_list, &sb->s_dirty);
167 }
168 }
169out:
170 spin_unlock(&inode_lock);
171}
172
173EXPORT_SYMBOL(__mark_inode_dirty);
174
175static int write_inode(struct inode *inode, int sync)
176{
177 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
178 return inode->i_sb->s_op->write_inode(inode, sync);
179 return 0;
180}
181
182/*
Andrew Morton6610a0b2007-10-16 23:30:32 -0700183 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
184 * furthest end of its superblock's dirty-inode list.
185 *
186 * Before stamping the inode's ->dirtied_when, we check to see whether it is
187 * already the most-recently-dirtied inode on the s_dirty list. If that is
188 * the case then the inode must have been redirtied while it was being written
189 * out and we don't reset its dirtied_when.
190 */
191static void redirty_tail(struct inode *inode)
192{
193 struct super_block *sb = inode->i_sb;
194
195 if (!list_empty(&sb->s_dirty)) {
196 struct inode *tail_inode;
197
198 tail_inode = list_entry(sb->s_dirty.next, struct inode, i_list);
Jeff Laytond2caa3c2009-04-02 16:56:37 -0700199 if (time_before(inode->dirtied_when,
Andrew Morton6610a0b2007-10-16 23:30:32 -0700200 tail_inode->dirtied_when))
201 inode->dirtied_when = jiffies;
202 }
203 list_move(&inode->i_list, &sb->s_dirty);
204}
205
206/*
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700207 * requeue inode for re-scanning after sb->s_io list is exhausted.
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700208 */
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700209static void requeue_io(struct inode *inode)
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700210{
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700211 list_move(&inode->i_list, &inode->i_sb->s_more_io);
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700212}
213
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700214static void inode_sync_complete(struct inode *inode)
215{
216 /*
217 * Prevent speculative execution through spin_unlock(&inode_lock);
218 */
219 smp_mb();
220 wake_up_bit(&inode->i_state, __I_SYNC);
221}
222
Jeff Laytond2caa3c2009-04-02 16:56:37 -0700223static bool inode_dirtied_after(struct inode *inode, unsigned long t)
224{
225 bool ret = time_after(inode->dirtied_when, t);
226#ifndef CONFIG_64BIT
227 /*
228 * For inodes being constantly redirtied, dirtied_when can get stuck.
229 * It _appears_ to be in the future, but is actually in distant past.
230 * This test is necessary to prevent such wrapped-around relative times
231 * from permanently stopping the whole pdflush writeback.
232 */
233 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
234#endif
235 return ret;
236}
237
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700238/*
Fengguang Wu2c136572007-10-16 23:30:39 -0700239 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
240 */
241static void move_expired_inodes(struct list_head *delaying_queue,
242 struct list_head *dispatch_queue,
243 unsigned long *older_than_this)
244{
245 while (!list_empty(delaying_queue)) {
246 struct inode *inode = list_entry(delaying_queue->prev,
247 struct inode, i_list);
248 if (older_than_this &&
Jeff Laytond2caa3c2009-04-02 16:56:37 -0700249 inode_dirtied_after(inode, *older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700250 break;
251 list_move(&inode->i_list, dispatch_queue);
252 }
253}
254
255/*
256 * Queue all expired dirty inodes for io, eldest first.
257 */
258static void queue_io(struct super_block *sb,
259 unsigned long *older_than_this)
260{
261 list_splice_init(&sb->s_more_io, sb->s_io.prev);
262 move_expired_inodes(&sb->s_dirty, &sb->s_io, older_than_this);
263}
264
Fengguang Wu08d8e972007-10-16 23:30:39 -0700265int sb_has_dirty_inodes(struct super_block *sb)
266{
267 return !list_empty(&sb->s_dirty) ||
268 !list_empty(&sb->s_io) ||
269 !list_empty(&sb->s_more_io);
270}
271EXPORT_SYMBOL(sb_has_dirty_inodes);
272
Fengguang Wu2c136572007-10-16 23:30:39 -0700273/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 * Write a single inode's dirty pages and inode data out to disk.
275 * If `wait' is set, wait on the writeout.
276 *
277 * The whole writeout design is quite complex and fragile. We want to avoid
278 * starvation of particular inodes when others are being redirtied, prevent
279 * livelocks, etc.
280 *
281 * Called under inode_lock.
282 */
283static int
284__sync_single_inode(struct inode *inode, struct writeback_control *wbc)
285{
286 unsigned dirty;
287 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 int wait = wbc->sync_mode == WB_SYNC_ALL;
289 int ret;
290
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700291 BUG_ON(inode->i_state & I_SYNC);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700292 WARN_ON(inode->i_state & I_NEW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700294 /* Set I_SYNC, reset I_DIRTY */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 dirty = inode->i_state & I_DIRTY;
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700296 inode->i_state |= I_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 inode->i_state &= ~I_DIRTY;
298
299 spin_unlock(&inode_lock);
300
301 ret = do_writepages(mapping, wbc);
302
303 /* Don't write the inode if only I_DIRTY_PAGES was set */
304 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
305 int err = write_inode(inode, wait);
306 if (ret == 0)
307 ret = err;
308 }
309
310 if (wait) {
311 int err = filemap_fdatawait(mapping);
312 if (ret == 0)
313 ret = err;
314 }
315
316 spin_lock(&inode_lock);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700317 WARN_ON(inode->i_state & I_NEW);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700318 inode->i_state &= ~I_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 if (!(inode->i_state & I_FREEING)) {
320 if (!(inode->i_state & I_DIRTY) &&
321 mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
322 /*
323 * We didn't write back all the pages. nfs_writepages()
324 * sometimes bales out without doing anything. Redirty
Fengguang Wu2c136572007-10-16 23:30:39 -0700325 * the inode; Move it from s_io onto s_more_io/s_dirty.
Andrew Morton1b43ef92007-10-16 23:30:35 -0700326 */
327 /*
328 * akpm: if the caller was the kupdate function we put
329 * this inode at the head of s_dirty so it gets first
330 * consideration. Otherwise, move it to the tail, for
331 * the reasons described there. I'm not really sure
332 * how much sense this makes. Presumably I had a good
333 * reasons for doing it this way, and I'd rather not
334 * muck with it at present.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 */
336 if (wbc->for_kupdate) {
337 /*
Fengguang Wu2c136572007-10-16 23:30:39 -0700338 * For the kupdate function we move the inode
339 * to s_more_io so it will get more writeout as
340 * soon as the queue becomes uncongested.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 */
342 inode->i_state |= I_DIRTY_PAGES;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800343 if (wbc->nr_to_write <= 0) {
344 /*
345 * slice used up: queue for next turn
346 */
347 requeue_io(inode);
348 } else {
349 /*
350 * somehow blocked: retry later
351 */
352 redirty_tail(inode);
353 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 } else {
355 /*
356 * Otherwise fully redirty the inode so that
357 * other inodes on this superblock will get some
358 * writeout. Otherwise heavy writing to one
359 * file would indefinitely suspend writeout of
360 * all the other files.
361 */
362 inode->i_state |= I_DIRTY_PAGES;
Andrew Morton1b43ef92007-10-16 23:30:35 -0700363 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 }
365 } else if (inode->i_state & I_DIRTY) {
366 /*
367 * Someone redirtied the inode while were writing back
368 * the pages.
369 */
Andrew Morton6610a0b2007-10-16 23:30:32 -0700370 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 } else if (atomic_read(&inode->i_count)) {
372 /*
373 * The inode is clean, inuse
374 */
375 list_move(&inode->i_list, &inode_in_use);
376 } else {
377 /*
378 * The inode is clean, unused
379 */
380 list_move(&inode->i_list, &inode_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 }
382 }
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700383 inode_sync_complete(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 return ret;
385}
386
387/*
Andrea Arcangeli7f04c262005-10-30 15:03:05 -0800388 * Write out an inode's dirty pages. Called under inode_lock. Either the
389 * caller has ref on the inode (either via __iget or via syscall against an fd)
390 * or the inode has I_WILL_FREE set (via generic_forget_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 */
392static int
Andrea Arcangeli7f04c262005-10-30 15:03:05 -0800393__writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394{
395 wait_queue_head_t *wqh;
396
Andrea Arcangeli7f04c262005-10-30 15:03:05 -0800397 if (!atomic_read(&inode->i_count))
Andrea Arcangeli659603e2005-10-31 14:08:54 -0800398 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
Andrea Arcangeli7f04c262005-10-30 15:03:05 -0800399 else
400 WARN_ON(inode->i_state & I_WILL_FREE);
401
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700402 if ((wbc->sync_mode != WB_SYNC_ALL) && (inode->i_state & I_SYNC)) {
Andrew Morton65cb9b42007-10-16 23:30:37 -0700403 /*
404 * We're skipping this inode because it's locked, and we're not
Fengguang Wu2c136572007-10-16 23:30:39 -0700405 * doing writeback-for-data-integrity. Move it to s_more_io so
406 * that writeback can proceed with the other inodes on s_io.
407 * We'll have another go at writing back this inode when we
408 * completed a full scan of s_io.
Andrew Morton65cb9b42007-10-16 23:30:37 -0700409 */
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700410 requeue_io(inode);
Qi Yong2d544562008-02-04 22:29:23 -0800411 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 }
413
414 /*
415 * It's a data-integrity sync. We must wait.
416 */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700417 if (inode->i_state & I_SYNC) {
418 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700420 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 spin_unlock(&inode_lock);
423 __wait_on_bit(wqh, &wq, inode_wait,
424 TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 spin_lock(&inode_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700426 } while (inode->i_state & I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 }
428 return __sync_single_inode(inode, wbc);
429}
430
431/*
432 * Write out a superblock's list of dirty inodes. A wait will be performed
433 * upon no inodes, all inodes or the final one, depending upon sync_mode.
434 *
435 * If older_than_this is non-NULL, then only write out inodes which
436 * had their first dirtying at a time earlier than *older_than_this.
437 *
Masatake YAMATO3e3cb642009-02-25 22:51:57 +0900438 * If we're a pdflush thread, then implement pdflush collision avoidance
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 * against the entire list.
440 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
442 * This function assumes that the blockdev superblock's inodes are backed by
443 * a variety of queues, so all inodes are searched. For other superblocks,
444 * assume that all inodes are backed by the same queue.
445 *
446 * FIXME: this linear search could get expensive with many fileystems. But
447 * how to fix? We need to go from an address_space to all inodes which share
448 * a queue with that address_space. (Easy: have a global "dirty superblocks"
449 * list).
450 *
451 * The inodes to be written are parked on sb->s_io. They are moved back onto
452 * sb->s_dirty as they are selected for writing. This way, none can be missed
453 * on the writer throttling path, and we get decent balancing between many
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700454 * throttled threads: we don't want them all piling up on inode_sync_wait.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 */
Artem Bityutskiy4ee6afd2008-05-07 21:01:30 +0300456void generic_sync_sb_inodes(struct super_block *sb,
457 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458{
459 const unsigned long start = jiffies; /* livelock avoidance */
Nick Piggin38f21972009-01-06 14:40:25 -0800460 int sync = wbc->sync_mode == WB_SYNC_ALL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
Hans Reiserae8547b2008-05-07 15:48:57 +0300462 spin_lock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 if (!wbc->for_kupdate || list_empty(&sb->s_io))
Fengguang Wu2c136572007-10-16 23:30:39 -0700464 queue_io(sb, wbc->older_than_this);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
466 while (!list_empty(&sb->s_io)) {
467 struct inode *inode = list_entry(sb->s_io.prev,
468 struct inode, i_list);
469 struct address_space *mapping = inode->i_mapping;
470 struct backing_dev_info *bdi = mapping->backing_dev_info;
471 long pages_skipped;
472
473 if (!bdi_cap_writeback_dirty(bdi)) {
Andrew Morton9852a0e72007-10-16 23:30:33 -0700474 redirty_tail(inode);
David Howells7b0de422006-08-29 19:06:07 +0100475 if (sb_is_blkdev_sb(sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 /*
477 * Dirty memory-backed blockdev: the ramdisk
478 * driver does this. Skip just this inode
479 */
480 continue;
481 }
482 /*
483 * Dirty memory-backed inode against a filesystem other
484 * than the kernel-internal bdev filesystem. Skip the
485 * entire superblock.
486 */
487 break;
488 }
489
Nick Piggin7ef0d732009-03-12 14:31:38 -0700490 if (inode->i_state & I_NEW) {
491 requeue_io(inode);
492 continue;
493 }
494
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 if (wbc->nonblocking && bdi_write_congested(bdi)) {
496 wbc->encountered_congestion = 1;
David Howells7b0de422006-08-29 19:06:07 +0100497 if (!sb_is_blkdev_sb(sb))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 break; /* Skip a congested fs */
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700499 requeue_io(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 continue; /* Skip a congested blockdev */
501 }
502
503 if (wbc->bdi && bdi != wbc->bdi) {
David Howells7b0de422006-08-29 19:06:07 +0100504 if (!sb_is_blkdev_sb(sb))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 break; /* fs has the wrong queue */
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700506 requeue_io(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 continue; /* blockdev has wrong queue */
508 }
509
Jeff Laytond2caa3c2009-04-02 16:56:37 -0700510 /*
511 * Was this inode dirtied after sync_sb_inodes was called?
512 * This keeps sync from extra jobs and livelock.
513 */
514 if (inode_dirtied_after(inode, start))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 break;
516
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 /* Is another pdflush already flushing this queue? */
518 if (current_is_pdflush() && !writeback_acquire(bdi))
519 break;
520
521 BUG_ON(inode->i_state & I_FREEING);
522 __iget(inode);
523 pages_skipped = wbc->pages_skipped;
524 __writeback_single_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 if (current_is_pdflush())
526 writeback_release(bdi);
527 if (wbc->pages_skipped != pages_skipped) {
528 /*
529 * writeback is not making progress due to locked
530 * buffers. Skip this inode for now.
531 */
Andrew Mortonf57b9b72007-10-16 23:30:34 -0700532 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 }
534 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 iput(inode);
OGAWA Hirofumi4ffc8442006-03-25 03:07:44 -0800536 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 spin_lock(&inode_lock);
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800538 if (wbc->nr_to_write <= 0) {
539 wbc->more_io = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800541 }
542 if (!list_empty(&sb->s_more_io))
543 wbc->more_io = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 }
Nick Piggin38f21972009-01-06 14:40:25 -0800545
546 if (sync) {
547 struct inode *inode, *old_inode = NULL;
548
549 /*
550 * Data integrity sync. Must wait for all pages under writeback,
551 * because there may have been pages dirtied before our sync
552 * call, but which had writeout started before we write it out.
553 * In which case, the inode may not be on the dirty list, but
554 * we still have to wait for that writeout.
555 */
556 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
557 struct address_space *mapping;
558
Wu Fengguangb6fac632009-04-02 16:56:34 -0700559 if (inode->i_state &
560 (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW))
Nick Piggin38f21972009-01-06 14:40:25 -0800561 continue;
562 mapping = inode->i_mapping;
563 if (mapping->nrpages == 0)
564 continue;
565 __iget(inode);
566 spin_unlock(&inode_lock);
567 /*
568 * We hold a reference to 'inode' so it couldn't have
569 * been removed from s_inodes list while we dropped the
570 * inode_lock. We cannot iput the inode now as we can
571 * be holding the last reference and we cannot iput it
572 * under inode_lock. So we keep the reference and iput
573 * it later.
574 */
575 iput(old_inode);
576 old_inode = inode;
577
578 filemap_fdatawait(mapping);
579
580 cond_resched();
581
582 spin_lock(&inode_lock);
583 }
584 spin_unlock(&inode_lock);
585 iput(old_inode);
586 } else
587 spin_unlock(&inode_lock);
588
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 return; /* Leave any unwritten inodes on s_io */
590}
Artem Bityutskiy4ee6afd2008-05-07 21:01:30 +0300591EXPORT_SYMBOL_GPL(generic_sync_sb_inodes);
592
593static void sync_sb_inodes(struct super_block *sb,
594 struct writeback_control *wbc)
595{
596 generic_sync_sb_inodes(sb, wbc);
597}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
599/*
600 * Start writeback of dirty pagecache data against all unlocked inodes.
601 *
602 * Note:
603 * We don't need to grab a reference to superblock here. If it has non-empty
604 * ->s_dirty it's hadn't been killed yet and kill_super() won't proceed
Fengguang Wu2c136572007-10-16 23:30:39 -0700605 * past sync_inodes_sb() until the ->s_dirty/s_io/s_more_io lists are all
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 * empty. Since __sync_single_inode() regains inode_lock before it finally moves
607 * inode from superblock lists we are OK.
608 *
609 * If `older_than_this' is non-zero then only flush inodes which have a
610 * flushtime older than *older_than_this.
611 *
612 * If `bdi' is non-zero then we will scan the first inode against each
613 * superblock until we find the matching ones. One group will be the dirty
614 * inodes against a filesystem. Then when we hit the dummy blockdev superblock,
615 * sync_sb_inodes will seekout the blockdev which matches `bdi'. Maybe not
616 * super-efficient but we're about to do a ton of I/O...
617 */
618void
619writeback_inodes(struct writeback_control *wbc)
620{
621 struct super_block *sb;
622
623 might_sleep();
624 spin_lock(&sb_lock);
625restart:
Akinobu Mita797074e2008-02-06 01:37:08 -0800626 list_for_each_entry_reverse(sb, &super_blocks, s_list) {
Fengguang Wu08d8e972007-10-16 23:30:39 -0700627 if (sb_has_dirty_inodes(sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 /* we're making our own get_super here */
629 sb->s_count++;
630 spin_unlock(&sb_lock);
631 /*
632 * If we can't get the readlock, there's no sense in
633 * waiting around, most of the time the FS is going to
634 * be unmounted by the time it is released.
635 */
636 if (down_read_trylock(&sb->s_umount)) {
Hans Reiserae8547b2008-05-07 15:48:57 +0300637 if (sb->s_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 sync_sb_inodes(sb, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 up_read(&sb->s_umount);
640 }
641 spin_lock(&sb_lock);
642 if (__put_super_and_need_restart(sb))
643 goto restart;
644 }
645 if (wbc->nr_to_write <= 0)
646 break;
647 }
648 spin_unlock(&sb_lock);
649}
650
651/*
652 * writeback and wait upon the filesystem's dirty inodes. The caller will
Nick Piggin4f5a99d2009-01-06 14:40:25 -0800653 * do this in two passes - one to write, and one to wait.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 *
655 * A finite limit is set on the number of pages which will be written.
656 * To prevent infinite livelock of sys_sync().
657 *
658 * We add in the number of potentially dirty inodes, because each inode write
659 * can dirty pagecache in the underlying blockdev.
660 */
661void sync_inodes_sb(struct super_block *sb, int wait)
662{
663 struct writeback_control wbc = {
Nick Piggin4f5a99d2009-01-06 14:40:25 -0800664 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700665 .range_start = 0,
666 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
Nick Piggin38f21972009-01-06 14:40:25 -0800669 if (!wait) {
670 unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY);
671 unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS);
672
673 wbc.nr_to_write = nr_dirty + nr_unstable +
674 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
675 } else
676 wbc.nr_to_write = LONG_MAX; /* doesn't actually matter */
677
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 sync_sb_inodes(sb, &wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679}
680
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681/**
Martin Waitz67be2dd2005-05-01 08:59:26 -0700682 * sync_inodes - writes all inodes to disk
683 * @wait: wait for completion
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 *
685 * sync_inodes() goes through each super block's dirty inode list, writes the
686 * inodes out, waits on the writeout and puts the inodes back on the normal
687 * list.
688 *
689 * This is for sys_sync(). fsync_dev() uses the same algorithm. The subtle
690 * part of the sync functions is that the blockdev "superblock" is processed
691 * last. This is because the write_inode() function of a typical fs will
692 * perform no I/O, but will mark buffers in the blockdev mapping as dirty.
693 * What we want to do is to perform all that dirtying first, and then write
694 * back all those inode blocks via the blockdev mapping in one sweep. So the
695 * additional (somewhat redundant) sync_blockdev() calls here are to make
696 * sure that really happens. Because if we call sync_inodes_sb(wait=1) with
697 * outstanding dirty inodes, the writeback goes block-at-a-time within the
698 * filesystem's write_inode(). This is extremely slow.
699 */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700700static void __sync_inodes(int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701{
702 struct super_block *sb;
703
Kirill Korotaev618f0632005-06-23 00:09:54 -0700704 spin_lock(&sb_lock);
705restart:
706 list_for_each_entry(sb, &super_blocks, s_list) {
Kirill Korotaev618f0632005-06-23 00:09:54 -0700707 sb->s_count++;
708 spin_unlock(&sb_lock);
709 down_read(&sb->s_umount);
710 if (sb->s_root) {
711 sync_inodes_sb(sb, wait);
712 sync_blockdev(sb->s_bdev);
713 }
714 up_read(&sb->s_umount);
715 spin_lock(&sb_lock);
716 if (__put_super_and_need_restart(sb))
717 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 }
Kirill Korotaev618f0632005-06-23 00:09:54 -0700719 spin_unlock(&sb_lock);
720}
721
722void sync_inodes(int wait)
723{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700724 __sync_inodes(0);
725
Nick Piggin856bf4d2009-01-06 14:40:26 -0800726 if (wait)
Kirill Korotaev618f0632005-06-23 00:09:54 -0700727 __sync_inodes(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728}
729
730/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -0800731 * write_inode_now - write an inode to disk
732 * @inode: inode to write to disk
733 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -0800735 * This function commits an inode to disk immediately if it is dirty. This is
736 * primarily needed by knfsd.
737 *
738 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740int write_inode_now(struct inode *inode, int sync)
741{
742 int ret;
743 struct writeback_control wbc = {
744 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -0800745 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700746 .range_start = 0,
747 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 };
749
750 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -0800751 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
753 might_sleep();
754 spin_lock(&inode_lock);
755 ret = __writeback_single_inode(inode, &wbc);
756 spin_unlock(&inode_lock);
757 if (sync)
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700758 inode_sync_wait(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 return ret;
760}
761EXPORT_SYMBOL(write_inode_now);
762
763/**
764 * sync_inode - write an inode and its pages to disk.
765 * @inode: the inode to sync
766 * @wbc: controls the writeback mode
767 *
768 * sync_inode() will write an inode and its pages to disk. It will also
769 * correctly update the inode on its superblock's dirty inode lists and will
770 * update inode->i_state.
771 *
772 * The caller must have a ref on the inode.
773 */
774int sync_inode(struct inode *inode, struct writeback_control *wbc)
775{
776 int ret;
777
778 spin_lock(&inode_lock);
779 ret = __writeback_single_inode(inode, wbc);
780 spin_unlock(&inode_lock);
781 return ret;
782}
783EXPORT_SYMBOL(sync_inode);
784
785/**
786 * generic_osync_inode - flush all dirty data for a given inode to disk
787 * @inode: inode to write
Martin Waitz67be2dd2005-05-01 08:59:26 -0700788 * @mapping: the address_space that should be flushed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 * @what: what to write and wait upon
790 *
791 * This can be called by file_write functions for files which have the
792 * O_SYNC flag set, to flush dirty writes to disk.
793 *
794 * @what is a bitmask, specifying which part of the inode's data should be
Randy Dunlapb8887e62005-11-07 01:01:07 -0800795 * written and waited upon.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 *
797 * OSYNC_DATA: i_mapping's dirty data
798 * OSYNC_METADATA: the buffers at i_mapping->private_list
799 * OSYNC_INODE: the inode itself
800 */
801
802int generic_osync_inode(struct inode *inode, struct address_space *mapping, int what)
803{
804 int err = 0;
805 int need_write_inode_now = 0;
806 int err2;
807
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 if (what & OSYNC_DATA)
809 err = filemap_fdatawrite(mapping);
810 if (what & (OSYNC_METADATA|OSYNC_DATA)) {
811 err2 = sync_mapping_buffers(mapping);
812 if (!err)
813 err = err2;
814 }
815 if (what & OSYNC_DATA) {
816 err2 = filemap_fdatawait(mapping);
817 if (!err)
818 err = err2;
819 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
821 spin_lock(&inode_lock);
822 if ((inode->i_state & I_DIRTY) &&
823 ((what & OSYNC_INODE) || (inode->i_state & I_DIRTY_DATASYNC)))
824 need_write_inode_now = 1;
825 spin_unlock(&inode_lock);
826
827 if (need_write_inode_now) {
828 err2 = write_inode_now(inode, 1);
829 if (!err)
830 err = err2;
831 }
832 else
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700833 inode_sync_wait(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834
835 return err;
836}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837EXPORT_SYMBOL(generic_osync_inode);