blob: c0076077d33874d0d63322d7c698ccbb04e4ed78 [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 *
11 * 10Apr2002 akpm@zip.com.au
12 * 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
28/**
29 * __mark_inode_dirty - internal function
30 * @inode: inode to mark
31 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
32 * Mark an inode as dirty. Callers should use mark_inode_dirty or
33 * mark_inode_dirty_sync.
34 *
35 * Put the inode on the super block's dirty list.
36 *
37 * CAREFUL! We mark it dirty unconditionally, but move it onto the
38 * dirty list only if it is hashed or if it refers to a blockdev.
39 * If it was not hashed, it will never be added to the dirty list
40 * even if it is later hashed, as it will have been marked dirty already.
41 *
42 * In short, make sure you hash any inodes _before_ you start marking
43 * them dirty.
44 *
45 * This function *must* be atomic for the I_DIRTY_PAGES case -
46 * set_page_dirty() is called under spinlock in several places.
47 *
48 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
49 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
50 * the kernel-internal blockdev inode represents the dirtying time of the
51 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
52 * page->mapping->host, so the page-dirtying time is recorded in the internal
53 * blockdev inode.
54 */
55void __mark_inode_dirty(struct inode *inode, int flags)
56{
57 struct super_block *sb = inode->i_sb;
58
59 /*
60 * Don't do this for I_DIRTY_PAGES - that doesn't actually
61 * dirty the inode itself
62 */
63 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
64 if (sb->s_op->dirty_inode)
65 sb->s_op->dirty_inode(inode);
66 }
67
68 /*
69 * make sure that changes are seen by all cpus before we test i_state
70 * -- mikulas
71 */
72 smp_mb();
73
74 /* avoid the locking if we can */
75 if ((inode->i_state & flags) == flags)
76 return;
77
78 if (unlikely(block_dump)) {
79 struct dentry *dentry = NULL;
80 const char *name = "?";
81
82 if (!list_empty(&inode->i_dentry)) {
83 dentry = list_entry(inode->i_dentry.next,
84 struct dentry, d_alias);
85 if (dentry && dentry->d_name.name)
86 name = (const char *) dentry->d_name.name;
87 }
88
89 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev"))
90 printk(KERN_DEBUG
91 "%s(%d): dirtied inode %lu (%s) on %s\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -070092 current->comm, task_pid_nr(current), inode->i_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 name, inode->i_sb->s_id);
94 }
95
96 spin_lock(&inode_lock);
97 if ((inode->i_state & flags) != flags) {
98 const int was_dirty = inode->i_state & I_DIRTY;
99
100 inode->i_state |= flags;
101
102 /*
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700103 * If the inode is being synced, just update its dirty state.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 * The unlocker will place the inode on the appropriate
105 * superblock list, based upon its state.
106 */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700107 if (inode->i_state & I_SYNC)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 goto out;
109
110 /*
111 * Only add valid (hashed) inodes to the superblock's
112 * dirty list. Add blockdev inodes as well.
113 */
114 if (!S_ISBLK(inode->i_mode)) {
115 if (hlist_unhashed(&inode->i_hash))
116 goto out;
117 }
118 if (inode->i_state & (I_FREEING|I_CLEAR))
119 goto out;
120
121 /*
Fengguang Wu2c136572007-10-16 23:30:39 -0700122 * If the inode was already on s_dirty/s_io/s_more_io, don't
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 * reposition it (that would break s_dirty time-ordering).
124 */
125 if (!was_dirty) {
126 inode->dirtied_when = jiffies;
127 list_move(&inode->i_list, &sb->s_dirty);
128 }
129 }
130out:
131 spin_unlock(&inode_lock);
132}
133
134EXPORT_SYMBOL(__mark_inode_dirty);
135
136static int write_inode(struct inode *inode, int sync)
137{
138 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
139 return inode->i_sb->s_op->write_inode(inode, sync);
140 return 0;
141}
142
143/*
Andrew Morton6610a0b2007-10-16 23:30:32 -0700144 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
145 * furthest end of its superblock's dirty-inode list.
146 *
147 * Before stamping the inode's ->dirtied_when, we check to see whether it is
148 * already the most-recently-dirtied inode on the s_dirty list. If that is
149 * the case then the inode must have been redirtied while it was being written
150 * out and we don't reset its dirtied_when.
151 */
152static void redirty_tail(struct inode *inode)
153{
154 struct super_block *sb = inode->i_sb;
155
156 if (!list_empty(&sb->s_dirty)) {
157 struct inode *tail_inode;
158
159 tail_inode = list_entry(sb->s_dirty.next, struct inode, i_list);
160 if (!time_after_eq(inode->dirtied_when,
161 tail_inode->dirtied_when))
162 inode->dirtied_when = jiffies;
163 }
164 list_move(&inode->i_list, &sb->s_dirty);
165}
166
167/*
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700168 * requeue inode for re-scanning after sb->s_io list is exhausted.
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700169 */
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700170static void requeue_io(struct inode *inode)
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700171{
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700172 list_move(&inode->i_list, &inode->i_sb->s_more_io);
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700173}
174
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700175static void inode_sync_complete(struct inode *inode)
176{
177 /*
178 * Prevent speculative execution through spin_unlock(&inode_lock);
179 */
180 smp_mb();
181 wake_up_bit(&inode->i_state, __I_SYNC);
182}
183
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700184/*
Fengguang Wu2c136572007-10-16 23:30:39 -0700185 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
186 */
187static void move_expired_inodes(struct list_head *delaying_queue,
188 struct list_head *dispatch_queue,
189 unsigned long *older_than_this)
190{
191 while (!list_empty(delaying_queue)) {
192 struct inode *inode = list_entry(delaying_queue->prev,
193 struct inode, i_list);
194 if (older_than_this &&
195 time_after(inode->dirtied_when, *older_than_this))
196 break;
197 list_move(&inode->i_list, dispatch_queue);
198 }
199}
200
201/*
202 * Queue all expired dirty inodes for io, eldest first.
203 */
204static void queue_io(struct super_block *sb,
205 unsigned long *older_than_this)
206{
207 list_splice_init(&sb->s_more_io, sb->s_io.prev);
208 move_expired_inodes(&sb->s_dirty, &sb->s_io, older_than_this);
209}
210
Fengguang Wu08d8e972007-10-16 23:30:39 -0700211int sb_has_dirty_inodes(struct super_block *sb)
212{
213 return !list_empty(&sb->s_dirty) ||
214 !list_empty(&sb->s_io) ||
215 !list_empty(&sb->s_more_io);
216}
217EXPORT_SYMBOL(sb_has_dirty_inodes);
218
Fengguang Wu2c136572007-10-16 23:30:39 -0700219/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 * Write a single inode's dirty pages and inode data out to disk.
221 * If `wait' is set, wait on the writeout.
222 *
223 * The whole writeout design is quite complex and fragile. We want to avoid
224 * starvation of particular inodes when others are being redirtied, prevent
225 * livelocks, etc.
226 *
227 * Called under inode_lock.
228 */
229static int
230__sync_single_inode(struct inode *inode, struct writeback_control *wbc)
231{
232 unsigned dirty;
233 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 int wait = wbc->sync_mode == WB_SYNC_ALL;
235 int ret;
236
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700237 BUG_ON(inode->i_state & I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700239 /* Set I_SYNC, reset I_DIRTY */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 dirty = inode->i_state & I_DIRTY;
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700241 inode->i_state |= I_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 inode->i_state &= ~I_DIRTY;
243
244 spin_unlock(&inode_lock);
245
246 ret = do_writepages(mapping, wbc);
247
248 /* Don't write the inode if only I_DIRTY_PAGES was set */
249 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
250 int err = write_inode(inode, wait);
251 if (ret == 0)
252 ret = err;
253 }
254
255 if (wait) {
256 int err = filemap_fdatawait(mapping);
257 if (ret == 0)
258 ret = err;
259 }
260
261 spin_lock(&inode_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700262 inode->i_state &= ~I_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 if (!(inode->i_state & I_FREEING)) {
264 if (!(inode->i_state & I_DIRTY) &&
265 mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
266 /*
267 * We didn't write back all the pages. nfs_writepages()
268 * sometimes bales out without doing anything. Redirty
Fengguang Wu2c136572007-10-16 23:30:39 -0700269 * the inode; Move it from s_io onto s_more_io/s_dirty.
Andrew Morton1b43ef92007-10-16 23:30:35 -0700270 */
271 /*
272 * akpm: if the caller was the kupdate function we put
273 * this inode at the head of s_dirty so it gets first
274 * consideration. Otherwise, move it to the tail, for
275 * the reasons described there. I'm not really sure
276 * how much sense this makes. Presumably I had a good
277 * reasons for doing it this way, and I'd rather not
278 * muck with it at present.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 */
280 if (wbc->for_kupdate) {
281 /*
Fengguang Wu2c136572007-10-16 23:30:39 -0700282 * For the kupdate function we move the inode
283 * to s_more_io so it will get more writeout as
284 * soon as the queue becomes uncongested.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 */
286 inode->i_state |= I_DIRTY_PAGES;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800287 if (wbc->nr_to_write <= 0) {
288 /*
289 * slice used up: queue for next turn
290 */
291 requeue_io(inode);
292 } else {
293 /*
294 * somehow blocked: retry later
295 */
296 redirty_tail(inode);
297 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 } else {
299 /*
300 * Otherwise fully redirty the inode so that
301 * other inodes on this superblock will get some
302 * writeout. Otherwise heavy writing to one
303 * file would indefinitely suspend writeout of
304 * all the other files.
305 */
306 inode->i_state |= I_DIRTY_PAGES;
Andrew Morton1b43ef92007-10-16 23:30:35 -0700307 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 }
309 } else if (inode->i_state & I_DIRTY) {
310 /*
311 * Someone redirtied the inode while were writing back
312 * the pages.
313 */
Andrew Morton6610a0b2007-10-16 23:30:32 -0700314 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 } else if (atomic_read(&inode->i_count)) {
316 /*
317 * The inode is clean, inuse
318 */
319 list_move(&inode->i_list, &inode_in_use);
320 } else {
321 /*
322 * The inode is clean, unused
323 */
324 list_move(&inode->i_list, &inode_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 }
326 }
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700327 inode_sync_complete(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 return ret;
329}
330
331/*
Andrea Arcangeli7f04c262005-10-30 15:03:05 -0800332 * Write out an inode's dirty pages. Called under inode_lock. Either the
333 * caller has ref on the inode (either via __iget or via syscall against an fd)
334 * or the inode has I_WILL_FREE set (via generic_forget_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 */
336static int
Andrea Arcangeli7f04c262005-10-30 15:03:05 -0800337__writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338{
339 wait_queue_head_t *wqh;
340
Andrea Arcangeli7f04c262005-10-30 15:03:05 -0800341 if (!atomic_read(&inode->i_count))
Andrea Arcangeli659603e2005-10-31 14:08:54 -0800342 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
Andrea Arcangeli7f04c262005-10-30 15:03:05 -0800343 else
344 WARN_ON(inode->i_state & I_WILL_FREE);
345
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700346 if ((wbc->sync_mode != WB_SYNC_ALL) && (inode->i_state & I_SYNC)) {
Andrew Morton65cb9b42007-10-16 23:30:37 -0700347 /*
348 * We're skipping this inode because it's locked, and we're not
Fengguang Wu2c136572007-10-16 23:30:39 -0700349 * doing writeback-for-data-integrity. Move it to s_more_io so
350 * that writeback can proceed with the other inodes on s_io.
351 * We'll have another go at writing back this inode when we
352 * completed a full scan of s_io.
Andrew Morton65cb9b42007-10-16 23:30:37 -0700353 */
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700354 requeue_io(inode);
Qi Yong2d544562008-02-04 22:29:23 -0800355 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 }
357
358 /*
359 * It's a data-integrity sync. We must wait.
360 */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700361 if (inode->i_state & I_SYNC) {
362 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700364 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 spin_unlock(&inode_lock);
367 __wait_on_bit(wqh, &wq, inode_wait,
368 TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 spin_lock(&inode_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700370 } while (inode->i_state & I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 }
372 return __sync_single_inode(inode, wbc);
373}
374
375/*
376 * Write out a superblock's list of dirty inodes. A wait will be performed
377 * upon no inodes, all inodes or the final one, depending upon sync_mode.
378 *
379 * If older_than_this is non-NULL, then only write out inodes which
380 * had their first dirtying at a time earlier than *older_than_this.
381 *
382 * If we're a pdlfush thread, then implement pdflush collision avoidance
383 * against the entire list.
384 *
385 * WB_SYNC_HOLD is a hack for sys_sync(): reattach the inode to sb->s_dirty so
386 * that it can be located for waiting on in __writeback_single_inode().
387 *
388 * Called under inode_lock.
389 *
390 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
391 * This function assumes that the blockdev superblock's inodes are backed by
392 * a variety of queues, so all inodes are searched. For other superblocks,
393 * assume that all inodes are backed by the same queue.
394 *
395 * FIXME: this linear search could get expensive with many fileystems. But
396 * how to fix? We need to go from an address_space to all inodes which share
397 * a queue with that address_space. (Easy: have a global "dirty superblocks"
398 * list).
399 *
400 * The inodes to be written are parked on sb->s_io. They are moved back onto
401 * sb->s_dirty as they are selected for writing. This way, none can be missed
402 * on the writer throttling path, and we get decent balancing between many
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700403 * throttled threads: we don't want them all piling up on inode_sync_wait.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 */
405static void
406sync_sb_inodes(struct super_block *sb, struct writeback_control *wbc)
407{
408 const unsigned long start = jiffies; /* livelock avoidance */
409
410 if (!wbc->for_kupdate || list_empty(&sb->s_io))
Fengguang Wu2c136572007-10-16 23:30:39 -0700411 queue_io(sb, wbc->older_than_this);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
413 while (!list_empty(&sb->s_io)) {
414 struct inode *inode = list_entry(sb->s_io.prev,
415 struct inode, i_list);
416 struct address_space *mapping = inode->i_mapping;
417 struct backing_dev_info *bdi = mapping->backing_dev_info;
418 long pages_skipped;
419
420 if (!bdi_cap_writeback_dirty(bdi)) {
Andrew Morton9852a0e72007-10-16 23:30:33 -0700421 redirty_tail(inode);
David Howells7b0de422006-08-29 19:06:07 +0100422 if (sb_is_blkdev_sb(sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 /*
424 * Dirty memory-backed blockdev: the ramdisk
425 * driver does this. Skip just this inode
426 */
427 continue;
428 }
429 /*
430 * Dirty memory-backed inode against a filesystem other
431 * than the kernel-internal bdev filesystem. Skip the
432 * entire superblock.
433 */
434 break;
435 }
436
437 if (wbc->nonblocking && bdi_write_congested(bdi)) {
438 wbc->encountered_congestion = 1;
David Howells7b0de422006-08-29 19:06:07 +0100439 if (!sb_is_blkdev_sb(sb))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 break; /* Skip a congested fs */
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700441 requeue_io(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 continue; /* Skip a congested blockdev */
443 }
444
445 if (wbc->bdi && bdi != wbc->bdi) {
David Howells7b0de422006-08-29 19:06:07 +0100446 if (!sb_is_blkdev_sb(sb))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 break; /* fs has the wrong queue */
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700448 requeue_io(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 continue; /* blockdev has wrong queue */
450 }
451
452 /* Was this inode dirtied after sync_sb_inodes was called? */
453 if (time_after(inode->dirtied_when, start))
454 break;
455
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 /* Is another pdflush already flushing this queue? */
457 if (current_is_pdflush() && !writeback_acquire(bdi))
458 break;
459
460 BUG_ON(inode->i_state & I_FREEING);
461 __iget(inode);
462 pages_skipped = wbc->pages_skipped;
463 __writeback_single_inode(inode, wbc);
464 if (wbc->sync_mode == WB_SYNC_HOLD) {
465 inode->dirtied_when = jiffies;
466 list_move(&inode->i_list, &sb->s_dirty);
467 }
468 if (current_is_pdflush())
469 writeback_release(bdi);
470 if (wbc->pages_skipped != pages_skipped) {
471 /*
472 * writeback is not making progress due to locked
473 * buffers. Skip this inode for now.
474 */
Andrew Mortonf57b9b72007-10-16 23:30:34 -0700475 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 }
477 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 iput(inode);
OGAWA Hirofumi4ffc8442006-03-25 03:07:44 -0800479 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 spin_lock(&inode_lock);
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800481 if (wbc->nr_to_write <= 0) {
482 wbc->more_io = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800484 }
485 if (!list_empty(&sb->s_more_io))
486 wbc->more_io = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 }
488 return; /* Leave any unwritten inodes on s_io */
489}
490
491/*
492 * Start writeback of dirty pagecache data against all unlocked inodes.
493 *
494 * Note:
495 * We don't need to grab a reference to superblock here. If it has non-empty
496 * ->s_dirty it's hadn't been killed yet and kill_super() won't proceed
Fengguang Wu2c136572007-10-16 23:30:39 -0700497 * past sync_inodes_sb() until the ->s_dirty/s_io/s_more_io lists are all
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 * empty. Since __sync_single_inode() regains inode_lock before it finally moves
499 * inode from superblock lists we are OK.
500 *
501 * If `older_than_this' is non-zero then only flush inodes which have a
502 * flushtime older than *older_than_this.
503 *
504 * If `bdi' is non-zero then we will scan the first inode against each
505 * superblock until we find the matching ones. One group will be the dirty
506 * inodes against a filesystem. Then when we hit the dummy blockdev superblock,
507 * sync_sb_inodes will seekout the blockdev which matches `bdi'. Maybe not
508 * super-efficient but we're about to do a ton of I/O...
509 */
510void
511writeback_inodes(struct writeback_control *wbc)
512{
513 struct super_block *sb;
514
515 might_sleep();
516 spin_lock(&sb_lock);
517restart:
Akinobu Mita797074e2008-02-06 01:37:08 -0800518 list_for_each_entry_reverse(sb, &super_blocks, s_list) {
Fengguang Wu08d8e972007-10-16 23:30:39 -0700519 if (sb_has_dirty_inodes(sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 /* we're making our own get_super here */
521 sb->s_count++;
522 spin_unlock(&sb_lock);
523 /*
524 * If we can't get the readlock, there's no sense in
525 * waiting around, most of the time the FS is going to
526 * be unmounted by the time it is released.
527 */
528 if (down_read_trylock(&sb->s_umount)) {
529 if (sb->s_root) {
530 spin_lock(&inode_lock);
531 sync_sb_inodes(sb, wbc);
532 spin_unlock(&inode_lock);
533 }
534 up_read(&sb->s_umount);
535 }
536 spin_lock(&sb_lock);
537 if (__put_super_and_need_restart(sb))
538 goto restart;
539 }
540 if (wbc->nr_to_write <= 0)
541 break;
542 }
543 spin_unlock(&sb_lock);
544}
545
546/*
547 * writeback and wait upon the filesystem's dirty inodes. The caller will
548 * do this in two passes - one to write, and one to wait. WB_SYNC_HOLD is
549 * used to park the written inodes on sb->s_dirty for the wait pass.
550 *
551 * A finite limit is set on the number of pages which will be written.
552 * To prevent infinite livelock of sys_sync().
553 *
554 * We add in the number of potentially dirty inodes, because each inode write
555 * can dirty pagecache in the underlying blockdev.
556 */
557void sync_inodes_sb(struct super_block *sb, int wait)
558{
559 struct writeback_control wbc = {
560 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_HOLD,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700561 .range_start = 0,
562 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 };
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -0700564 unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY);
Christoph Lameterfd39fc82006-06-30 01:55:40 -0700565 unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
567 wbc.nr_to_write = nr_dirty + nr_unstable +
568 (inodes_stat.nr_inodes - inodes_stat.nr_unused) +
569 nr_dirty + nr_unstable;
570 wbc.nr_to_write += wbc.nr_to_write / 2; /* Bit more for luck */
571 spin_lock(&inode_lock);
572 sync_sb_inodes(sb, &wbc);
573 spin_unlock(&inode_lock);
574}
575
576/*
577 * Rather lame livelock avoidance.
578 */
579static void set_sb_syncing(int val)
580{
581 struct super_block *sb;
582 spin_lock(&sb_lock);
Akinobu Mita797074e2008-02-06 01:37:08 -0800583 list_for_each_entry_reverse(sb, &super_blocks, s_list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 sb->s_syncing = val;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 spin_unlock(&sb_lock);
586}
587
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588/**
Martin Waitz67be2dd2005-05-01 08:59:26 -0700589 * sync_inodes - writes all inodes to disk
590 * @wait: wait for completion
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 *
592 * sync_inodes() goes through each super block's dirty inode list, writes the
593 * inodes out, waits on the writeout and puts the inodes back on the normal
594 * list.
595 *
596 * This is for sys_sync(). fsync_dev() uses the same algorithm. The subtle
597 * part of the sync functions is that the blockdev "superblock" is processed
598 * last. This is because the write_inode() function of a typical fs will
599 * perform no I/O, but will mark buffers in the blockdev mapping as dirty.
600 * What we want to do is to perform all that dirtying first, and then write
601 * back all those inode blocks via the blockdev mapping in one sweep. So the
602 * additional (somewhat redundant) sync_blockdev() calls here are to make
603 * sure that really happens. Because if we call sync_inodes_sb(wait=1) with
604 * outstanding dirty inodes, the writeback goes block-at-a-time within the
605 * filesystem's write_inode(). This is extremely slow.
606 */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700607static void __sync_inodes(int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608{
609 struct super_block *sb;
610
Kirill Korotaev618f0632005-06-23 00:09:54 -0700611 spin_lock(&sb_lock);
612restart:
613 list_for_each_entry(sb, &super_blocks, s_list) {
614 if (sb->s_syncing)
615 continue;
616 sb->s_syncing = 1;
617 sb->s_count++;
618 spin_unlock(&sb_lock);
619 down_read(&sb->s_umount);
620 if (sb->s_root) {
621 sync_inodes_sb(sb, wait);
622 sync_blockdev(sb->s_bdev);
623 }
624 up_read(&sb->s_umount);
625 spin_lock(&sb_lock);
626 if (__put_super_and_need_restart(sb))
627 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 }
Kirill Korotaev618f0632005-06-23 00:09:54 -0700629 spin_unlock(&sb_lock);
630}
631
632void sync_inodes(int wait)
633{
634 set_sb_syncing(0);
635 __sync_inodes(0);
636
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 if (wait) {
638 set_sb_syncing(0);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700639 __sync_inodes(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 }
641}
642
643/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -0800644 * write_inode_now - write an inode to disk
645 * @inode: inode to write to disk
646 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -0800648 * This function commits an inode to disk immediately if it is dirty. This is
649 * primarily needed by knfsd.
650 *
651 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653int write_inode_now(struct inode *inode, int sync)
654{
655 int ret;
656 struct writeback_control wbc = {
657 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -0800658 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700659 .range_start = 0,
660 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 };
662
663 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -0800664 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665
666 might_sleep();
667 spin_lock(&inode_lock);
668 ret = __writeback_single_inode(inode, &wbc);
669 spin_unlock(&inode_lock);
670 if (sync)
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700671 inode_sync_wait(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 return ret;
673}
674EXPORT_SYMBOL(write_inode_now);
675
676/**
677 * sync_inode - write an inode and its pages to disk.
678 * @inode: the inode to sync
679 * @wbc: controls the writeback mode
680 *
681 * sync_inode() will write an inode and its pages to disk. It will also
682 * correctly update the inode on its superblock's dirty inode lists and will
683 * update inode->i_state.
684 *
685 * The caller must have a ref on the inode.
686 */
687int sync_inode(struct inode *inode, struct writeback_control *wbc)
688{
689 int ret;
690
691 spin_lock(&inode_lock);
692 ret = __writeback_single_inode(inode, wbc);
693 spin_unlock(&inode_lock);
694 return ret;
695}
696EXPORT_SYMBOL(sync_inode);
697
698/**
699 * generic_osync_inode - flush all dirty data for a given inode to disk
700 * @inode: inode to write
Martin Waitz67be2dd2005-05-01 08:59:26 -0700701 * @mapping: the address_space that should be flushed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 * @what: what to write and wait upon
703 *
704 * This can be called by file_write functions for files which have the
705 * O_SYNC flag set, to flush dirty writes to disk.
706 *
707 * @what is a bitmask, specifying which part of the inode's data should be
Randy Dunlapb8887e62005-11-07 01:01:07 -0800708 * written and waited upon.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 *
710 * OSYNC_DATA: i_mapping's dirty data
711 * OSYNC_METADATA: the buffers at i_mapping->private_list
712 * OSYNC_INODE: the inode itself
713 */
714
715int generic_osync_inode(struct inode *inode, struct address_space *mapping, int what)
716{
717 int err = 0;
718 int need_write_inode_now = 0;
719 int err2;
720
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 if (what & OSYNC_DATA)
722 err = filemap_fdatawrite(mapping);
723 if (what & (OSYNC_METADATA|OSYNC_DATA)) {
724 err2 = sync_mapping_buffers(mapping);
725 if (!err)
726 err = err2;
727 }
728 if (what & OSYNC_DATA) {
729 err2 = filemap_fdatawait(mapping);
730 if (!err)
731 err = err2;
732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733
734 spin_lock(&inode_lock);
735 if ((inode->i_state & I_DIRTY) &&
736 ((what & OSYNC_INODE) || (inode->i_state & I_DIRTY_DATASYNC)))
737 need_write_inode_now = 1;
738 spin_unlock(&inode_lock);
739
740 if (need_write_inode_now) {
741 err2 = write_inode_now(inode, 1);
742 if (!err)
743 err = err2;
744 }
745 else
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700746 inode_sync_wait(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
748 return err;
749}
750
751EXPORT_SYMBOL(generic_osync_inode);
752
753/**
754 * writeback_acquire: attempt to get exclusive writeback access to a device
755 * @bdi: the device's backing_dev_info structure
756 *
757 * It is a waste of resources to have more than one pdflush thread blocked on
758 * a single request queue. Exclusion at the request_queue level is obtained
759 * via a flag in the request_queue's backing_dev_info.state.
760 *
761 * Non-request_queue-backed address_spaces will share default_backing_dev_info,
762 * unless they implement their own. Which is somewhat inefficient, as this
763 * may prevent concurrent writeback against multiple devices.
764 */
765int writeback_acquire(struct backing_dev_info *bdi)
766{
767 return !test_and_set_bit(BDI_pdflush, &bdi->state);
768}
769
770/**
771 * writeback_in_progress: determine whether there is writeback in progress
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 * @bdi: the device's backing_dev_info structure.
Randy Dunlapb8887e62005-11-07 01:01:07 -0800773 *
774 * Determine whether there is writeback in progress against a backing device.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 */
776int writeback_in_progress(struct backing_dev_info *bdi)
777{
778 return test_bit(BDI_pdflush, &bdi->state);
779}
780
781/**
782 * writeback_release: relinquish exclusive writeback access against a device.
783 * @bdi: the device's backing_dev_info structure
784 */
785void writeback_release(struct backing_dev_info *bdi)
786{
787 BUG_ON(!writeback_in_progress(bdi));
788 clear_bit(BDI_pdflush, &bdi->state);
789}