1 |
62 |
marcus.erl |
/*
|
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>
|
17 |
|
|
#include <linux/module.h>
|
18 |
|
|
#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>
|
26 |
|
|
#include "internal.h"
|
27 |
|
|
|
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 |
|
|
*/
|
55 |
|
|
void __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",
|
92 |
|
|
current->comm, task_pid_nr(current), inode->i_ino,
|
93 |
|
|
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 |
|
|
/*
|
103 |
|
|
* If the inode is being synced, just update its dirty state.
|
104 |
|
|
* The unlocker will place the inode on the appropriate
|
105 |
|
|
* superblock list, based upon its state.
|
106 |
|
|
*/
|
107 |
|
|
if (inode->i_state & I_SYNC)
|
108 |
|
|
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 |
|
|
/*
|
122 |
|
|
* If the inode was already on s_dirty/s_io/s_more_io, don't
|
123 |
|
|
* 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 |
|
|
}
|
130 |
|
|
out:
|
131 |
|
|
spin_unlock(&inode_lock);
|
132 |
|
|
}
|
133 |
|
|
|
134 |
|
|
EXPORT_SYMBOL(__mark_inode_dirty);
|
135 |
|
|
|
136 |
|
|
static 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 |
|
|
/*
|
144 |
|
|
* 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 |
|
|
*/
|
152 |
|
|
static 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 |
|
|
/*
|
168 |
|
|
* requeue inode for re-scanning after sb->s_io list is exhausted.
|
169 |
|
|
*/
|
170 |
|
|
static void requeue_io(struct inode *inode)
|
171 |
|
|
{
|
172 |
|
|
list_move(&inode->i_list, &inode->i_sb->s_more_io);
|
173 |
|
|
}
|
174 |
|
|
|
175 |
|
|
static 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 |
|
|
|
184 |
|
|
/*
|
185 |
|
|
* Move expired dirty inodes from @delaying_queue to @dispatch_queue.
|
186 |
|
|
*/
|
187 |
|
|
static 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 |
|
|
*/
|
204 |
|
|
static 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 |
|
|
|
211 |
|
|
int 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 |
|
|
}
|
217 |
|
|
EXPORT_SYMBOL(sb_has_dirty_inodes);
|
218 |
|
|
|
219 |
|
|
/*
|
220 |
|
|
* 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 |
|
|
*/
|
229 |
|
|
static int
|
230 |
|
|
__sync_single_inode(struct inode *inode, struct writeback_control *wbc)
|
231 |
|
|
{
|
232 |
|
|
unsigned dirty;
|
233 |
|
|
struct address_space *mapping = inode->i_mapping;
|
234 |
|
|
int wait = wbc->sync_mode == WB_SYNC_ALL;
|
235 |
|
|
int ret;
|
236 |
|
|
|
237 |
|
|
BUG_ON(inode->i_state & I_SYNC);
|
238 |
|
|
|
239 |
|
|
/* Set I_SYNC, reset I_DIRTY */
|
240 |
|
|
dirty = inode->i_state & I_DIRTY;
|
241 |
|
|
inode->i_state |= I_SYNC;
|
242 |
|
|
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);
|
262 |
|
|
inode->i_state &= ~I_SYNC;
|
263 |
|
|
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
|
269 |
|
|
* the inode; Move it from s_io onto s_more_io/s_dirty.
|
270 |
|
|
*/
|
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.
|
279 |
|
|
*/
|
280 |
|
|
if (wbc->for_kupdate) {
|
281 |
|
|
/*
|
282 |
|
|
* 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.
|
285 |
|
|
*/
|
286 |
|
|
inode->i_state |= I_DIRTY_PAGES;
|
287 |
|
|
requeue_io(inode);
|
288 |
|
|
} else {
|
289 |
|
|
/*
|
290 |
|
|
* Otherwise fully redirty the inode so that
|
291 |
|
|
* other inodes on this superblock will get some
|
292 |
|
|
* writeout. Otherwise heavy writing to one
|
293 |
|
|
* file would indefinitely suspend writeout of
|
294 |
|
|
* all the other files.
|
295 |
|
|
*/
|
296 |
|
|
inode->i_state |= I_DIRTY_PAGES;
|
297 |
|
|
redirty_tail(inode);
|
298 |
|
|
}
|
299 |
|
|
} else if (inode->i_state & I_DIRTY) {
|
300 |
|
|
/*
|
301 |
|
|
* Someone redirtied the inode while were writing back
|
302 |
|
|
* the pages.
|
303 |
|
|
*/
|
304 |
|
|
redirty_tail(inode);
|
305 |
|
|
} else if (atomic_read(&inode->i_count)) {
|
306 |
|
|
/*
|
307 |
|
|
* The inode is clean, inuse
|
308 |
|
|
*/
|
309 |
|
|
list_move(&inode->i_list, &inode_in_use);
|
310 |
|
|
} else {
|
311 |
|
|
/*
|
312 |
|
|
* The inode is clean, unused
|
313 |
|
|
*/
|
314 |
|
|
list_move(&inode->i_list, &inode_unused);
|
315 |
|
|
}
|
316 |
|
|
}
|
317 |
|
|
inode_sync_complete(inode);
|
318 |
|
|
return ret;
|
319 |
|
|
}
|
320 |
|
|
|
321 |
|
|
/*
|
322 |
|
|
* Write out an inode's dirty pages. Called under inode_lock. Either the
|
323 |
|
|
* caller has ref on the inode (either via __iget or via syscall against an fd)
|
324 |
|
|
* or the inode has I_WILL_FREE set (via generic_forget_inode)
|
325 |
|
|
*/
|
326 |
|
|
static int
|
327 |
|
|
__writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
|
328 |
|
|
{
|
329 |
|
|
wait_queue_head_t *wqh;
|
330 |
|
|
|
331 |
|
|
if (!atomic_read(&inode->i_count))
|
332 |
|
|
WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
|
333 |
|
|
else
|
334 |
|
|
WARN_ON(inode->i_state & I_WILL_FREE);
|
335 |
|
|
|
336 |
|
|
if ((wbc->sync_mode != WB_SYNC_ALL) && (inode->i_state & I_SYNC)) {
|
337 |
|
|
struct address_space *mapping = inode->i_mapping;
|
338 |
|
|
int ret;
|
339 |
|
|
|
340 |
|
|
/*
|
341 |
|
|
* We're skipping this inode because it's locked, and we're not
|
342 |
|
|
* doing writeback-for-data-integrity. Move it to s_more_io so
|
343 |
|
|
* that writeback can proceed with the other inodes on s_io.
|
344 |
|
|
* We'll have another go at writing back this inode when we
|
345 |
|
|
* completed a full scan of s_io.
|
346 |
|
|
*/
|
347 |
|
|
requeue_io(inode);
|
348 |
|
|
|
349 |
|
|
/*
|
350 |
|
|
* Even if we don't actually write the inode itself here,
|
351 |
|
|
* we can at least start some of the data writeout..
|
352 |
|
|
*/
|
353 |
|
|
spin_unlock(&inode_lock);
|
354 |
|
|
ret = do_writepages(mapping, wbc);
|
355 |
|
|
spin_lock(&inode_lock);
|
356 |
|
|
return ret;
|
357 |
|
|
}
|
358 |
|
|
|
359 |
|
|
/*
|
360 |
|
|
* It's a data-integrity sync. We must wait.
|
361 |
|
|
*/
|
362 |
|
|
if (inode->i_state & I_SYNC) {
|
363 |
|
|
DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
|
364 |
|
|
|
365 |
|
|
wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
|
366 |
|
|
do {
|
367 |
|
|
spin_unlock(&inode_lock);
|
368 |
|
|
__wait_on_bit(wqh, &wq, inode_wait,
|
369 |
|
|
TASK_UNINTERRUPTIBLE);
|
370 |
|
|
spin_lock(&inode_lock);
|
371 |
|
|
} while (inode->i_state & I_SYNC);
|
372 |
|
|
}
|
373 |
|
|
return __sync_single_inode(inode, wbc);
|
374 |
|
|
}
|
375 |
|
|
|
376 |
|
|
/*
|
377 |
|
|
* Write out a superblock's list of dirty inodes. A wait will be performed
|
378 |
|
|
* upon no inodes, all inodes or the final one, depending upon sync_mode.
|
379 |
|
|
*
|
380 |
|
|
* If older_than_this is non-NULL, then only write out inodes which
|
381 |
|
|
* had their first dirtying at a time earlier than *older_than_this.
|
382 |
|
|
*
|
383 |
|
|
* If we're a pdlfush thread, then implement pdflush collision avoidance
|
384 |
|
|
* against the entire list.
|
385 |
|
|
*
|
386 |
|
|
* WB_SYNC_HOLD is a hack for sys_sync(): reattach the inode to sb->s_dirty so
|
387 |
|
|
* that it can be located for waiting on in __writeback_single_inode().
|
388 |
|
|
*
|
389 |
|
|
* Called under inode_lock.
|
390 |
|
|
*
|
391 |
|
|
* If `bdi' is non-zero then we're being asked to writeback a specific queue.
|
392 |
|
|
* This function assumes that the blockdev superblock's inodes are backed by
|
393 |
|
|
* a variety of queues, so all inodes are searched. For other superblocks,
|
394 |
|
|
* assume that all inodes are backed by the same queue.
|
395 |
|
|
*
|
396 |
|
|
* FIXME: this linear search could get expensive with many fileystems. But
|
397 |
|
|
* how to fix? We need to go from an address_space to all inodes which share
|
398 |
|
|
* a queue with that address_space. (Easy: have a global "dirty superblocks"
|
399 |
|
|
* list).
|
400 |
|
|
*
|
401 |
|
|
* The inodes to be written are parked on sb->s_io. They are moved back onto
|
402 |
|
|
* sb->s_dirty as they are selected for writing. This way, none can be missed
|
403 |
|
|
* on the writer throttling path, and we get decent balancing between many
|
404 |
|
|
* throttled threads: we don't want them all piling up on inode_sync_wait.
|
405 |
|
|
*/
|
406 |
|
|
static void
|
407 |
|
|
sync_sb_inodes(struct super_block *sb, struct writeback_control *wbc)
|
408 |
|
|
{
|
409 |
|
|
const unsigned long start = jiffies; /* livelock avoidance */
|
410 |
|
|
|
411 |
|
|
if (!wbc->for_kupdate || list_empty(&sb->s_io))
|
412 |
|
|
queue_io(sb, wbc->older_than_this);
|
413 |
|
|
|
414 |
|
|
while (!list_empty(&sb->s_io)) {
|
415 |
|
|
struct inode *inode = list_entry(sb->s_io.prev,
|
416 |
|
|
struct inode, i_list);
|
417 |
|
|
struct address_space *mapping = inode->i_mapping;
|
418 |
|
|
struct backing_dev_info *bdi = mapping->backing_dev_info;
|
419 |
|
|
long pages_skipped;
|
420 |
|
|
|
421 |
|
|
if (!bdi_cap_writeback_dirty(bdi)) {
|
422 |
|
|
redirty_tail(inode);
|
423 |
|
|
if (sb_is_blkdev_sb(sb)) {
|
424 |
|
|
/*
|
425 |
|
|
* Dirty memory-backed blockdev: the ramdisk
|
426 |
|
|
* driver does this. Skip just this inode
|
427 |
|
|
*/
|
428 |
|
|
continue;
|
429 |
|
|
}
|
430 |
|
|
/*
|
431 |
|
|
* Dirty memory-backed inode against a filesystem other
|
432 |
|
|
* than the kernel-internal bdev filesystem. Skip the
|
433 |
|
|
* entire superblock.
|
434 |
|
|
*/
|
435 |
|
|
break;
|
436 |
|
|
}
|
437 |
|
|
|
438 |
|
|
if (wbc->nonblocking && bdi_write_congested(bdi)) {
|
439 |
|
|
wbc->encountered_congestion = 1;
|
440 |
|
|
if (!sb_is_blkdev_sb(sb))
|
441 |
|
|
break; /* Skip a congested fs */
|
442 |
|
|
requeue_io(inode);
|
443 |
|
|
continue; /* Skip a congested blockdev */
|
444 |
|
|
}
|
445 |
|
|
|
446 |
|
|
if (wbc->bdi && bdi != wbc->bdi) {
|
447 |
|
|
if (!sb_is_blkdev_sb(sb))
|
448 |
|
|
break; /* fs has the wrong queue */
|
449 |
|
|
requeue_io(inode);
|
450 |
|
|
continue; /* blockdev has wrong queue */
|
451 |
|
|
}
|
452 |
|
|
|
453 |
|
|
/* Was this inode dirtied after sync_sb_inodes was called? */
|
454 |
|
|
if (time_after(inode->dirtied_when, start))
|
455 |
|
|
break;
|
456 |
|
|
|
457 |
|
|
/* Is another pdflush already flushing this queue? */
|
458 |
|
|
if (current_is_pdflush() && !writeback_acquire(bdi))
|
459 |
|
|
break;
|
460 |
|
|
|
461 |
|
|
BUG_ON(inode->i_state & I_FREEING);
|
462 |
|
|
__iget(inode);
|
463 |
|
|
pages_skipped = wbc->pages_skipped;
|
464 |
|
|
__writeback_single_inode(inode, wbc);
|
465 |
|
|
if (wbc->sync_mode == WB_SYNC_HOLD) {
|
466 |
|
|
inode->dirtied_when = jiffies;
|
467 |
|
|
list_move(&inode->i_list, &sb->s_dirty);
|
468 |
|
|
}
|
469 |
|
|
if (current_is_pdflush())
|
470 |
|
|
writeback_release(bdi);
|
471 |
|
|
if (wbc->pages_skipped != pages_skipped) {
|
472 |
|
|
/*
|
473 |
|
|
* writeback is not making progress due to locked
|
474 |
|
|
* buffers. Skip this inode for now.
|
475 |
|
|
*/
|
476 |
|
|
redirty_tail(inode);
|
477 |
|
|
}
|
478 |
|
|
spin_unlock(&inode_lock);
|
479 |
|
|
iput(inode);
|
480 |
|
|
cond_resched();
|
481 |
|
|
spin_lock(&inode_lock);
|
482 |
|
|
if (wbc->nr_to_write <= 0)
|
483 |
|
|
break;
|
484 |
|
|
}
|
485 |
|
|
return; /* Leave any unwritten inodes on s_io */
|
486 |
|
|
}
|
487 |
|
|
|
488 |
|
|
/*
|
489 |
|
|
* Start writeback of dirty pagecache data against all unlocked inodes.
|
490 |
|
|
*
|
491 |
|
|
* Note:
|
492 |
|
|
* We don't need to grab a reference to superblock here. If it has non-empty
|
493 |
|
|
* ->s_dirty it's hadn't been killed yet and kill_super() won't proceed
|
494 |
|
|
* past sync_inodes_sb() until the ->s_dirty/s_io/s_more_io lists are all
|
495 |
|
|
* empty. Since __sync_single_inode() regains inode_lock before it finally moves
|
496 |
|
|
* inode from superblock lists we are OK.
|
497 |
|
|
*
|
498 |
|
|
* If `older_than_this' is non-zero then only flush inodes which have a
|
499 |
|
|
* flushtime older than *older_than_this.
|
500 |
|
|
*
|
501 |
|
|
* If `bdi' is non-zero then we will scan the first inode against each
|
502 |
|
|
* superblock until we find the matching ones. One group will be the dirty
|
503 |
|
|
* inodes against a filesystem. Then when we hit the dummy blockdev superblock,
|
504 |
|
|
* sync_sb_inodes will seekout the blockdev which matches `bdi'. Maybe not
|
505 |
|
|
* super-efficient but we're about to do a ton of I/O...
|
506 |
|
|
*/
|
507 |
|
|
void
|
508 |
|
|
writeback_inodes(struct writeback_control *wbc)
|
509 |
|
|
{
|
510 |
|
|
struct super_block *sb;
|
511 |
|
|
|
512 |
|
|
might_sleep();
|
513 |
|
|
spin_lock(&sb_lock);
|
514 |
|
|
restart:
|
515 |
|
|
sb = sb_entry(super_blocks.prev);
|
516 |
|
|
for (; sb != sb_entry(&super_blocks); sb = sb_entry(sb->s_list.prev)) {
|
517 |
|
|
if (sb_has_dirty_inodes(sb)) {
|
518 |
|
|
/* we're making our own get_super here */
|
519 |
|
|
sb->s_count++;
|
520 |
|
|
spin_unlock(&sb_lock);
|
521 |
|
|
/*
|
522 |
|
|
* If we can't get the readlock, there's no sense in
|
523 |
|
|
* waiting around, most of the time the FS is going to
|
524 |
|
|
* be unmounted by the time it is released.
|
525 |
|
|
*/
|
526 |
|
|
if (down_read_trylock(&sb->s_umount)) {
|
527 |
|
|
if (sb->s_root) {
|
528 |
|
|
spin_lock(&inode_lock);
|
529 |
|
|
sync_sb_inodes(sb, wbc);
|
530 |
|
|
spin_unlock(&inode_lock);
|
531 |
|
|
}
|
532 |
|
|
up_read(&sb->s_umount);
|
533 |
|
|
}
|
534 |
|
|
spin_lock(&sb_lock);
|
535 |
|
|
if (__put_super_and_need_restart(sb))
|
536 |
|
|
goto restart;
|
537 |
|
|
}
|
538 |
|
|
if (wbc->nr_to_write <= 0)
|
539 |
|
|
break;
|
540 |
|
|
}
|
541 |
|
|
spin_unlock(&sb_lock);
|
542 |
|
|
}
|
543 |
|
|
|
544 |
|
|
/*
|
545 |
|
|
* writeback and wait upon the filesystem's dirty inodes. The caller will
|
546 |
|
|
* do this in two passes - one to write, and one to wait. WB_SYNC_HOLD is
|
547 |
|
|
* used to park the written inodes on sb->s_dirty for the wait pass.
|
548 |
|
|
*
|
549 |
|
|
* A finite limit is set on the number of pages which will be written.
|
550 |
|
|
* To prevent infinite livelock of sys_sync().
|
551 |
|
|
*
|
552 |
|
|
* We add in the number of potentially dirty inodes, because each inode write
|
553 |
|
|
* can dirty pagecache in the underlying blockdev.
|
554 |
|
|
*/
|
555 |
|
|
void sync_inodes_sb(struct super_block *sb, int wait)
|
556 |
|
|
{
|
557 |
|
|
struct writeback_control wbc = {
|
558 |
|
|
.sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_HOLD,
|
559 |
|
|
.range_start = 0,
|
560 |
|
|
.range_end = LLONG_MAX,
|
561 |
|
|
};
|
562 |
|
|
unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY);
|
563 |
|
|
unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS);
|
564 |
|
|
|
565 |
|
|
wbc.nr_to_write = nr_dirty + nr_unstable +
|
566 |
|
|
(inodes_stat.nr_inodes - inodes_stat.nr_unused) +
|
567 |
|
|
nr_dirty + nr_unstable;
|
568 |
|
|
wbc.nr_to_write += wbc.nr_to_write / 2; /* Bit more for luck */
|
569 |
|
|
spin_lock(&inode_lock);
|
570 |
|
|
sync_sb_inodes(sb, &wbc);
|
571 |
|
|
spin_unlock(&inode_lock);
|
572 |
|
|
}
|
573 |
|
|
|
574 |
|
|
/*
|
575 |
|
|
* Rather lame livelock avoidance.
|
576 |
|
|
*/
|
577 |
|
|
static void set_sb_syncing(int val)
|
578 |
|
|
{
|
579 |
|
|
struct super_block *sb;
|
580 |
|
|
spin_lock(&sb_lock);
|
581 |
|
|
sb = sb_entry(super_blocks.prev);
|
582 |
|
|
for (; sb != sb_entry(&super_blocks); sb = sb_entry(sb->s_list.prev)) {
|
583 |
|
|
sb->s_syncing = val;
|
584 |
|
|
}
|
585 |
|
|
spin_unlock(&sb_lock);
|
586 |
|
|
}
|
587 |
|
|
|
588 |
|
|
/**
|
589 |
|
|
* sync_inodes - writes all inodes to disk
|
590 |
|
|
* @wait: wait for completion
|
591 |
|
|
*
|
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 |
|
|
*/
|
607 |
|
|
static void __sync_inodes(int wait)
|
608 |
|
|
{
|
609 |
|
|
struct super_block *sb;
|
610 |
|
|
|
611 |
|
|
spin_lock(&sb_lock);
|
612 |
|
|
restart:
|
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;
|
628 |
|
|
}
|
629 |
|
|
spin_unlock(&sb_lock);
|
630 |
|
|
}
|
631 |
|
|
|
632 |
|
|
void sync_inodes(int wait)
|
633 |
|
|
{
|
634 |
|
|
set_sb_syncing(0);
|
635 |
|
|
__sync_inodes(0);
|
636 |
|
|
|
637 |
|
|
if (wait) {
|
638 |
|
|
set_sb_syncing(0);
|
639 |
|
|
__sync_inodes(1);
|
640 |
|
|
}
|
641 |
|
|
}
|
642 |
|
|
|
643 |
|
|
/**
|
644 |
|
|
* 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
|
647 |
|
|
*
|
648 |
|
|
* 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.
|
652 |
|
|
*/
|
653 |
|
|
int write_inode_now(struct inode *inode, int sync)
|
654 |
|
|
{
|
655 |
|
|
int ret;
|
656 |
|
|
struct writeback_control wbc = {
|
657 |
|
|
.nr_to_write = LONG_MAX,
|
658 |
|
|
.sync_mode = WB_SYNC_ALL,
|
659 |
|
|
.range_start = 0,
|
660 |
|
|
.range_end = LLONG_MAX,
|
661 |
|
|
};
|
662 |
|
|
|
663 |
|
|
if (!mapping_cap_writeback_dirty(inode->i_mapping))
|
664 |
|
|
wbc.nr_to_write = 0;
|
665 |
|
|
|
666 |
|
|
might_sleep();
|
667 |
|
|
spin_lock(&inode_lock);
|
668 |
|
|
ret = __writeback_single_inode(inode, &wbc);
|
669 |
|
|
spin_unlock(&inode_lock);
|
670 |
|
|
if (sync)
|
671 |
|
|
inode_sync_wait(inode);
|
672 |
|
|
return ret;
|
673 |
|
|
}
|
674 |
|
|
EXPORT_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 |
|
|
*/
|
687 |
|
|
int 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 |
|
|
}
|
696 |
|
|
EXPORT_SYMBOL(sync_inode);
|
697 |
|
|
|
698 |
|
|
/**
|
699 |
|
|
* generic_osync_inode - flush all dirty data for a given inode to disk
|
700 |
|
|
* @inode: inode to write
|
701 |
|
|
* @mapping: the address_space that should be flushed
|
702 |
|
|
* @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
|
708 |
|
|
* written and waited upon.
|
709 |
|
|
*
|
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 |
|
|
|
715 |
|
|
int 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 |
|
|
|
721 |
|
|
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 |
|
|
}
|
733 |
|
|
|
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
|
746 |
|
|
inode_sync_wait(inode);
|
747 |
|
|
|
748 |
|
|
return err;
|
749 |
|
|
}
|
750 |
|
|
|
751 |
|
|
EXPORT_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 |
|
|
*/
|
765 |
|
|
int 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
|
772 |
|
|
* @bdi: the device's backing_dev_info structure.
|
773 |
|
|
*
|
774 |
|
|
* Determine whether there is writeback in progress against a backing device.
|
775 |
|
|
*/
|
776 |
|
|
int 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 |
|
|
*/
|
785 |
|
|
void writeback_release(struct backing_dev_info *bdi)
|
786 |
|
|
{
|
787 |
|
|
BUG_ON(!writeback_in_progress(bdi));
|
788 |
|
|
clear_bit(BDI_pdflush, &bdi->state);
|
789 |
|
|
}
|