blob: 40308e98c6a44f9763354b375ba4c51bb569607a [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
Nick Piggin4195f732009-05-28 09:01:15 +020067static noinline void block_dump___mark_inode_dirty(struct inode *inode)
68{
69 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
70 struct dentry *dentry;
71 const char *name = "?";
72
73 dentry = d_find_alias(inode);
74 if (dentry) {
75 spin_lock(&dentry->d_lock);
76 name = (const char *) dentry->d_name.name;
77 }
78 printk(KERN_DEBUG
79 "%s(%d): dirtied inode %lu (%s) on %s\n",
80 current->comm, task_pid_nr(current), inode->i_ino,
81 name, inode->i_sb->s_id);
82 if (dentry) {
83 spin_unlock(&dentry->d_lock);
84 dput(dentry);
85 }
86 }
87}
88
Linus Torvalds1da177e2005-04-16 15:20:36 -070089/**
90 * __mark_inode_dirty - internal function
91 * @inode: inode to mark
92 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
93 * Mark an inode as dirty. Callers should use mark_inode_dirty or
94 * mark_inode_dirty_sync.
95 *
96 * Put the inode on the super block's dirty list.
97 *
98 * CAREFUL! We mark it dirty unconditionally, but move it onto the
99 * dirty list only if it is hashed or if it refers to a blockdev.
100 * If it was not hashed, it will never be added to the dirty list
101 * even if it is later hashed, as it will have been marked dirty already.
102 *
103 * In short, make sure you hash any inodes _before_ you start marking
104 * them dirty.
105 *
106 * This function *must* be atomic for the I_DIRTY_PAGES case -
107 * set_page_dirty() is called under spinlock in several places.
108 *
109 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
110 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
111 * the kernel-internal blockdev inode represents the dirtying time of the
112 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
113 * page->mapping->host, so the page-dirtying time is recorded in the internal
114 * blockdev inode.
115 */
116void __mark_inode_dirty(struct inode *inode, int flags)
117{
118 struct super_block *sb = inode->i_sb;
119
120 /*
121 * Don't do this for I_DIRTY_PAGES - that doesn't actually
122 * dirty the inode itself
123 */
124 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
125 if (sb->s_op->dirty_inode)
126 sb->s_op->dirty_inode(inode);
127 }
128
129 /*
130 * make sure that changes are seen by all cpus before we test i_state
131 * -- mikulas
132 */
133 smp_mb();
134
135 /* avoid the locking if we can */
136 if ((inode->i_state & flags) == flags)
137 return;
138
Nick Piggin4195f732009-05-28 09:01:15 +0200139 if (unlikely(block_dump))
140 block_dump___mark_inode_dirty(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
142 spin_lock(&inode_lock);
143 if ((inode->i_state & flags) != flags) {
144 const int was_dirty = inode->i_state & I_DIRTY;
145
146 inode->i_state |= flags;
147
148 /*
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700149 * If the inode is being synced, just update its dirty state.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 * The unlocker will place the inode on the appropriate
151 * superblock list, based upon its state.
152 */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700153 if (inode->i_state & I_SYNC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 goto out;
155
156 /*
157 * Only add valid (hashed) inodes to the superblock's
158 * dirty list. Add blockdev inodes as well.
159 */
160 if (!S_ISBLK(inode->i_mode)) {
161 if (hlist_unhashed(&inode->i_hash))
162 goto out;
163 }
164 if (inode->i_state & (I_FREEING|I_CLEAR))
165 goto out;
166
167 /*
Fengguang Wu2c136572007-10-16 23:30:39 -0700168 * If the inode was already on s_dirty/s_io/s_more_io, don't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 * reposition it (that would break s_dirty time-ordering).
170 */
171 if (!was_dirty) {
172 inode->dirtied_when = jiffies;
173 list_move(&inode->i_list, &sb->s_dirty);
174 }
175 }
176out:
177 spin_unlock(&inode_lock);
178}
179
180EXPORT_SYMBOL(__mark_inode_dirty);
181
182static int write_inode(struct inode *inode, int sync)
183{
184 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
185 return inode->i_sb->s_op->write_inode(inode, sync);
186 return 0;
187}
188
189/*
Andrew Morton6610a0b2007-10-16 23:30:32 -0700190 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
191 * furthest end of its superblock's dirty-inode list.
192 *
193 * Before stamping the inode's ->dirtied_when, we check to see whether it is
194 * already the most-recently-dirtied inode on the s_dirty list. If that is
195 * the case then the inode must have been redirtied while it was being written
196 * out and we don't reset its dirtied_when.
197 */
198static void redirty_tail(struct inode *inode)
199{
200 struct super_block *sb = inode->i_sb;
201
202 if (!list_empty(&sb->s_dirty)) {
203 struct inode *tail_inode;
204
205 tail_inode = list_entry(sb->s_dirty.next, struct inode, i_list);
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700206 if (time_before(inode->dirtied_when,
Andrew Morton6610a0b2007-10-16 23:30:32 -0700207 tail_inode->dirtied_when))
208 inode->dirtied_when = jiffies;
209 }
210 list_move(&inode->i_list, &sb->s_dirty);
211}
212
213/*
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700214 * requeue inode for re-scanning after sb->s_io list is exhausted.
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700215 */
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700216static void requeue_io(struct inode *inode)
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700217{
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700218 list_move(&inode->i_list, &inode->i_sb->s_more_io);
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700219}
220
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700221static void inode_sync_complete(struct inode *inode)
222{
223 /*
224 * Prevent speculative execution through spin_unlock(&inode_lock);
225 */
226 smp_mb();
227 wake_up_bit(&inode->i_state, __I_SYNC);
228}
229
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700230static bool inode_dirtied_after(struct inode *inode, unsigned long t)
231{
232 bool ret = time_after(inode->dirtied_when, t);
233#ifndef CONFIG_64BIT
234 /*
235 * For inodes being constantly redirtied, dirtied_when can get stuck.
236 * It _appears_ to be in the future, but is actually in distant past.
237 * This test is necessary to prevent such wrapped-around relative times
238 * from permanently stopping the whole pdflush writeback.
239 */
240 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
241#endif
242 return ret;
243}
244
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700245/*
Fengguang Wu2c136572007-10-16 23:30:39 -0700246 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
247 */
248static void move_expired_inodes(struct list_head *delaying_queue,
249 struct list_head *dispatch_queue,
250 unsigned long *older_than_this)
251{
252 while (!list_empty(delaying_queue)) {
253 struct inode *inode = list_entry(delaying_queue->prev,
254 struct inode, i_list);
255 if (older_than_this &&
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700256 inode_dirtied_after(inode, *older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700257 break;
258 list_move(&inode->i_list, dispatch_queue);
259 }
260}
261
262/*
263 * Queue all expired dirty inodes for io, eldest first.
264 */
265static void queue_io(struct super_block *sb,
266 unsigned long *older_than_this)
267{
268 list_splice_init(&sb->s_more_io, sb->s_io.prev);
269 move_expired_inodes(&sb->s_dirty, &sb->s_io, older_than_this);
270}
271
Fengguang Wu08d8e972007-10-16 23:30:39 -0700272int sb_has_dirty_inodes(struct super_block *sb)
273{
274 return !list_empty(&sb->s_dirty) ||
275 !list_empty(&sb->s_io) ||
276 !list_empty(&sb->s_more_io);
277}
278EXPORT_SYMBOL(sb_has_dirty_inodes);
279
Fengguang Wu2c136572007-10-16 23:30:39 -0700280/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 * Write a single inode's dirty pages and inode data out to disk.
282 * If `wait' is set, wait on the writeout.
283 *
284 * The whole writeout design is quite complex and fragile. We want to avoid
285 * starvation of particular inodes when others are being redirtied, prevent
286 * livelocks, etc.
287 *
288 * Called under inode_lock.
289 */
290static int
291__sync_single_inode(struct inode *inode, struct writeback_control *wbc)
292{
293 unsigned dirty;
294 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 int wait = wbc->sync_mode == WB_SYNC_ALL;
296 int ret;
297
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700298 BUG_ON(inode->i_state & I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700300 /* Set I_SYNC, reset I_DIRTY */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 dirty = inode->i_state & I_DIRTY;
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700302 inode->i_state |= I_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 inode->i_state &= ~I_DIRTY;
304
305 spin_unlock(&inode_lock);
306
307 ret = do_writepages(mapping, wbc);
308
309 /* Don't write the inode if only I_DIRTY_PAGES was set */
310 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
311 int err = write_inode(inode, wait);
312 if (ret == 0)
313 ret = err;
314 }
315
316 if (wait) {
317 int err = filemap_fdatawait(mapping);
318 if (ret == 0)
319 ret = err;
320 }
321
322 spin_lock(&inode_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700323 inode->i_state &= ~I_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 if (!(inode->i_state & I_FREEING)) {
325 if (!(inode->i_state & I_DIRTY) &&
326 mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
327 /*
328 * We didn't write back all the pages. nfs_writepages()
329 * sometimes bales out without doing anything. Redirty
Fengguang Wu2c136572007-10-16 23:30:39 -0700330 * the inode; Move it from s_io onto s_more_io/s_dirty.
Andrew Morton1b43ef92007-10-16 23:30:35 -0700331 */
332 /*
333 * akpm: if the caller was the kupdate function we put
334 * this inode at the head of s_dirty so it gets first
335 * consideration. Otherwise, move it to the tail, for
336 * the reasons described there. I'm not really sure
337 * how much sense this makes. Presumably I had a good
338 * reasons for doing it this way, and I'd rather not
339 * muck with it at present.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 */
341 if (wbc->for_kupdate) {
342 /*
Fengguang Wu2c136572007-10-16 23:30:39 -0700343 * For the kupdate function we move the inode
344 * to s_more_io so it will get more writeout as
345 * soon as the queue becomes uncongested.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 */
347 inode->i_state |= I_DIRTY_PAGES;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800348 if (wbc->nr_to_write <= 0) {
349 /*
350 * slice used up: queue for next turn
351 */
352 requeue_io(inode);
353 } else {
354 /*
355 * somehow blocked: retry later
356 */
357 redirty_tail(inode);
358 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 } else {
360 /*
361 * Otherwise fully redirty the inode so that
362 * other inodes on this superblock will get some
363 * writeout. Otherwise heavy writing to one
364 * file would indefinitely suspend writeout of
365 * all the other files.
366 */
367 inode->i_state |= I_DIRTY_PAGES;
Andrew Morton1b43ef92007-10-16 23:30:35 -0700368 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 }
370 } else if (inode->i_state & I_DIRTY) {
371 /*
372 * Someone redirtied the inode while were writing back
373 * the pages.
374 */
Andrew Morton6610a0b2007-10-16 23:30:32 -0700375 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 } else if (atomic_read(&inode->i_count)) {
377 /*
378 * The inode is clean, inuse
379 */
380 list_move(&inode->i_list, &inode_in_use);
381 } else {
382 /*
383 * The inode is clean, unused
384 */
385 list_move(&inode->i_list, &inode_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 }
387 }
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700388 inode_sync_complete(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 return ret;
390}
391
392/*
Andrea Arcangeli7f04c262005-10-30 15:03:05 -0800393 * Write out an inode's dirty pages. Called under inode_lock. Either the
394 * caller has ref on the inode (either via __iget or via syscall against an fd)
395 * or the inode has I_WILL_FREE set (via generic_forget_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 */
397static int
Andrea Arcangeli7f04c262005-10-30 15:03:05 -0800398__writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399{
400 wait_queue_head_t *wqh;
401
Andrea Arcangeli7f04c262005-10-30 15:03:05 -0800402 if (!atomic_read(&inode->i_count))
Andrea Arcangeli659603e2005-10-31 14:08:54 -0800403 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
Andrea Arcangeli7f04c262005-10-30 15:03:05 -0800404 else
405 WARN_ON(inode->i_state & I_WILL_FREE);
406
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700407 if ((wbc->sync_mode != WB_SYNC_ALL) && (inode->i_state & I_SYNC)) {
Andrew Morton65cb9b42007-10-16 23:30:37 -0700408 /*
409 * We're skipping this inode because it's locked, and we're not
Fengguang Wu2c136572007-10-16 23:30:39 -0700410 * doing writeback-for-data-integrity. Move it to s_more_io so
411 * that writeback can proceed with the other inodes on s_io.
412 * We'll have another go at writing back this inode when we
413 * completed a full scan of s_io.
Andrew Morton65cb9b42007-10-16 23:30:37 -0700414 */
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700415 requeue_io(inode);
Qi Yong2d544562008-02-04 22:29:23 -0800416 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 }
418
419 /*
420 * It's a data-integrity sync. We must wait.
421 */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700422 if (inode->i_state & I_SYNC) {
423 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700425 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 spin_unlock(&inode_lock);
428 __wait_on_bit(wqh, &wq, inode_wait,
429 TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 spin_lock(&inode_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700431 } while (inode->i_state & I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 }
433 return __sync_single_inode(inode, wbc);
434}
435
436/*
437 * Write out a superblock's list of dirty inodes. A wait will be performed
438 * upon no inodes, all inodes or the final one, depending upon sync_mode.
439 *
440 * If older_than_this is non-NULL, then only write out inodes which
441 * had their first dirtying at a time earlier than *older_than_this.
442 *
Masatake YAMATO3e3cb642009-02-25 22:51:57 +0900443 * If we're a pdflush thread, then implement pdflush collision avoidance
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 * against the entire list.
445 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
447 * This function assumes that the blockdev superblock's inodes are backed by
448 * a variety of queues, so all inodes are searched. For other superblocks,
449 * assume that all inodes are backed by the same queue.
450 *
451 * FIXME: this linear search could get expensive with many fileystems. But
452 * how to fix? We need to go from an address_space to all inodes which share
453 * a queue with that address_space. (Easy: have a global "dirty superblocks"
454 * list).
455 *
456 * The inodes to be written are parked on sb->s_io. They are moved back onto
457 * sb->s_dirty as they are selected for writing. This way, none can be missed
458 * on the writer throttling path, and we get decent balancing between many
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700459 * throttled threads: we don't want them all piling up on inode_sync_wait.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 */
Artem Bityutskiy4ee6afd2008-05-07 21:01:30 +0300461void generic_sync_sb_inodes(struct super_block *sb,
462 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463{
464 const unsigned long start = jiffies; /* livelock avoidance */
Nick Piggin38f21972009-01-06 14:40:25 -0800465 int sync = wbc->sync_mode == WB_SYNC_ALL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
Hans Reiserae8547b2008-05-07 15:48:57 +0300467 spin_lock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 if (!wbc->for_kupdate || list_empty(&sb->s_io))
Fengguang Wu2c136572007-10-16 23:30:39 -0700469 queue_io(sb, wbc->older_than_this);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470
471 while (!list_empty(&sb->s_io)) {
472 struct inode *inode = list_entry(sb->s_io.prev,
473 struct inode, i_list);
474 struct address_space *mapping = inode->i_mapping;
475 struct backing_dev_info *bdi = mapping->backing_dev_info;
476 long pages_skipped;
477
478 if (!bdi_cap_writeback_dirty(bdi)) {
Andrew Morton9852a0e72007-10-16 23:30:33 -0700479 redirty_tail(inode);
David Howells7b0de422006-08-29 19:06:07 +0100480 if (sb_is_blkdev_sb(sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 /*
482 * Dirty memory-backed blockdev: the ramdisk
483 * driver does this. Skip just this inode
484 */
485 continue;
486 }
487 /*
488 * Dirty memory-backed inode against a filesystem other
489 * than the kernel-internal bdev filesystem. Skip the
490 * entire superblock.
491 */
492 break;
493 }
494
Nick Piggin7ef0d732009-03-12 14:31:38 -0700495 if (inode->i_state & I_NEW) {
496 requeue_io(inode);
497 continue;
498 }
499
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 if (wbc->nonblocking && bdi_write_congested(bdi)) {
501 wbc->encountered_congestion = 1;
David Howells7b0de422006-08-29 19:06:07 +0100502 if (!sb_is_blkdev_sb(sb))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 break; /* Skip a congested fs */
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700504 requeue_io(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 continue; /* Skip a congested blockdev */
506 }
507
508 if (wbc->bdi && bdi != wbc->bdi) {
David Howells7b0de422006-08-29 19:06:07 +0100509 if (!sb_is_blkdev_sb(sb))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 break; /* fs has the wrong queue */
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700511 requeue_io(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 continue; /* blockdev has wrong queue */
513 }
514
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700515 /*
516 * Was this inode dirtied after sync_sb_inodes was called?
517 * This keeps sync from extra jobs and livelock.
518 */
519 if (inode_dirtied_after(inode, start))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 break;
521
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 /* Is another pdflush already flushing this queue? */
523 if (current_is_pdflush() && !writeback_acquire(bdi))
524 break;
525
526 BUG_ON(inode->i_state & I_FREEING);
527 __iget(inode);
528 pages_skipped = wbc->pages_skipped;
529 __writeback_single_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 if (current_is_pdflush())
531 writeback_release(bdi);
532 if (wbc->pages_skipped != pages_skipped) {
533 /*
534 * writeback is not making progress due to locked
535 * buffers. Skip this inode for now.
536 */
Andrew Mortonf57b9b72007-10-16 23:30:34 -0700537 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 }
539 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 iput(inode);
OGAWA Hirofumi4ffc8442006-03-25 03:07:44 -0800541 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 spin_lock(&inode_lock);
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800543 if (wbc->nr_to_write <= 0) {
544 wbc->more_io = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800546 }
547 if (!list_empty(&sb->s_more_io))
548 wbc->more_io = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 }
Nick Piggin38f21972009-01-06 14:40:25 -0800550
551 if (sync) {
552 struct inode *inode, *old_inode = NULL;
553
554 /*
555 * Data integrity sync. Must wait for all pages under writeback,
556 * because there may have been pages dirtied before our sync
557 * call, but which had writeout started before we write it out.
558 * In which case, the inode may not be on the dirty list, but
559 * we still have to wait for that writeout.
560 */
561 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
562 struct address_space *mapping;
563
Wu Fengguangb6fac632009-04-02 16:56:34 -0700564 if (inode->i_state &
565 (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW))
Nick Piggin38f21972009-01-06 14:40:25 -0800566 continue;
567 mapping = inode->i_mapping;
568 if (mapping->nrpages == 0)
569 continue;
570 __iget(inode);
571 spin_unlock(&inode_lock);
572 /*
573 * We hold a reference to 'inode' so it couldn't have
574 * been removed from s_inodes list while we dropped the
575 * inode_lock. We cannot iput the inode now as we can
576 * be holding the last reference and we cannot iput it
577 * under inode_lock. So we keep the reference and iput
578 * it later.
579 */
580 iput(old_inode);
581 old_inode = inode;
582
583 filemap_fdatawait(mapping);
584
585 cond_resched();
586
587 spin_lock(&inode_lock);
588 }
589 spin_unlock(&inode_lock);
590 iput(old_inode);
591 } else
592 spin_unlock(&inode_lock);
593
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 return; /* Leave any unwritten inodes on s_io */
595}
Artem Bityutskiy4ee6afd2008-05-07 21:01:30 +0300596EXPORT_SYMBOL_GPL(generic_sync_sb_inodes);
597
598static void sync_sb_inodes(struct super_block *sb,
599 struct writeback_control *wbc)
600{
601 generic_sync_sb_inodes(sb, wbc);
602}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603
604/*
605 * Start writeback of dirty pagecache data against all unlocked inodes.
606 *
607 * Note:
608 * We don't need to grab a reference to superblock here. If it has non-empty
609 * ->s_dirty it's hadn't been killed yet and kill_super() won't proceed
Fengguang Wu2c136572007-10-16 23:30:39 -0700610 * past sync_inodes_sb() until the ->s_dirty/s_io/s_more_io lists are all
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 * empty. Since __sync_single_inode() regains inode_lock before it finally moves
612 * inode from superblock lists we are OK.
613 *
614 * If `older_than_this' is non-zero then only flush inodes which have a
615 * flushtime older than *older_than_this.
616 *
617 * If `bdi' is non-zero then we will scan the first inode against each
618 * superblock until we find the matching ones. One group will be the dirty
619 * inodes against a filesystem. Then when we hit the dummy blockdev superblock,
620 * sync_sb_inodes will seekout the blockdev which matches `bdi'. Maybe not
621 * super-efficient but we're about to do a ton of I/O...
622 */
623void
624writeback_inodes(struct writeback_control *wbc)
625{
626 struct super_block *sb;
627
628 might_sleep();
629 spin_lock(&sb_lock);
630restart:
Akinobu Mita797074e2008-02-06 01:37:08 -0800631 list_for_each_entry_reverse(sb, &super_blocks, s_list) {
Fengguang Wu08d8e972007-10-16 23:30:39 -0700632 if (sb_has_dirty_inodes(sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 /* we're making our own get_super here */
634 sb->s_count++;
635 spin_unlock(&sb_lock);
636 /*
637 * If we can't get the readlock, there's no sense in
638 * waiting around, most of the time the FS is going to
639 * be unmounted by the time it is released.
640 */
641 if (down_read_trylock(&sb->s_umount)) {
Hans Reiserae8547b2008-05-07 15:48:57 +0300642 if (sb->s_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 sync_sb_inodes(sb, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 up_read(&sb->s_umount);
645 }
646 spin_lock(&sb_lock);
647 if (__put_super_and_need_restart(sb))
648 goto restart;
649 }
650 if (wbc->nr_to_write <= 0)
651 break;
652 }
653 spin_unlock(&sb_lock);
654}
655
656/*
657 * writeback and wait upon the filesystem's dirty inodes. The caller will
Nick Piggin4f5a99d2009-01-06 14:40:25 -0800658 * do this in two passes - one to write, and one to wait.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 *
660 * A finite limit is set on the number of pages which will be written.
661 * To prevent infinite livelock of sys_sync().
662 *
663 * We add in the number of potentially dirty inodes, because each inode write
664 * can dirty pagecache in the underlying blockdev.
665 */
666void sync_inodes_sb(struct super_block *sb, int wait)
667{
668 struct writeback_control wbc = {
Nick Piggin4f5a99d2009-01-06 14:40:25 -0800669 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700670 .range_start = 0,
671 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
Nick Piggin38f21972009-01-06 14:40:25 -0800674 if (!wait) {
675 unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY);
676 unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS);
677
678 wbc.nr_to_write = nr_dirty + nr_unstable +
679 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
680 } else
681 wbc.nr_to_write = LONG_MAX; /* doesn't actually matter */
682
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 sync_sb_inodes(sb, &wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684}
685
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -0800687 * write_inode_now - write an inode to disk
688 * @inode: inode to write to disk
689 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -0800691 * This function commits an inode to disk immediately if it is dirty. This is
692 * primarily needed by knfsd.
693 *
694 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696int write_inode_now(struct inode *inode, int sync)
697{
698 int ret;
699 struct writeback_control wbc = {
700 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -0800701 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700702 .range_start = 0,
703 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 };
705
706 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -0800707 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708
709 might_sleep();
710 spin_lock(&inode_lock);
711 ret = __writeback_single_inode(inode, &wbc);
712 spin_unlock(&inode_lock);
713 if (sync)
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700714 inode_sync_wait(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 return ret;
716}
717EXPORT_SYMBOL(write_inode_now);
718
719/**
720 * sync_inode - write an inode and its pages to disk.
721 * @inode: the inode to sync
722 * @wbc: controls the writeback mode
723 *
724 * sync_inode() will write an inode and its pages to disk. It will also
725 * correctly update the inode on its superblock's dirty inode lists and will
726 * update inode->i_state.
727 *
728 * The caller must have a ref on the inode.
729 */
730int sync_inode(struct inode *inode, struct writeback_control *wbc)
731{
732 int ret;
733
734 spin_lock(&inode_lock);
735 ret = __writeback_single_inode(inode, wbc);
736 spin_unlock(&inode_lock);
737 return ret;
738}
739EXPORT_SYMBOL(sync_inode);
740
741/**
742 * generic_osync_inode - flush all dirty data for a given inode to disk
743 * @inode: inode to write
Martin Waitz67be2dd2005-05-01 08:59:26 -0700744 * @mapping: the address_space that should be flushed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 * @what: what to write and wait upon
746 *
747 * This can be called by file_write functions for files which have the
748 * O_SYNC flag set, to flush dirty writes to disk.
749 *
750 * @what is a bitmask, specifying which part of the inode's data should be
Randy Dunlapb8887e62005-11-07 01:01:07 -0800751 * written and waited upon.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 *
753 * OSYNC_DATA: i_mapping's dirty data
754 * OSYNC_METADATA: the buffers at i_mapping->private_list
755 * OSYNC_INODE: the inode itself
756 */
757
758int generic_osync_inode(struct inode *inode, struct address_space *mapping, int what)
759{
760 int err = 0;
761 int need_write_inode_now = 0;
762 int err2;
763
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 if (what & OSYNC_DATA)
765 err = filemap_fdatawrite(mapping);
766 if (what & (OSYNC_METADATA|OSYNC_DATA)) {
767 err2 = sync_mapping_buffers(mapping);
768 if (!err)
769 err = err2;
770 }
771 if (what & OSYNC_DATA) {
772 err2 = filemap_fdatawait(mapping);
773 if (!err)
774 err = err2;
775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776
777 spin_lock(&inode_lock);
778 if ((inode->i_state & I_DIRTY) &&
779 ((what & OSYNC_INODE) || (inode->i_state & I_DIRTY_DATASYNC)))
780 need_write_inode_now = 1;
781 spin_unlock(&inode_lock);
782
783 if (need_write_inode_now) {
784 err2 = write_inode_now(inode, 1);
785 if (!err)
786 err = err2;
787 }
788 else
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700789 inode_sync_wait(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
791 return err;
792}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793EXPORT_SYMBOL(generic_osync_inode);