Doug / smarc-fsl-linux-kernel

Download as
Browse Code ยป

Commit 4195f73d1329e49727bcceb028e58cb38376c2b0

Authored by Nick Piggin
Committed by Al Viro
1 parent 545b9fd3d7
Exists in master and in 7 other branches imx_3.0.35_4.1.0, imx_3.10.17_1.0.1_ga, imx_3.10.53_1.1.0_ga, imx_3.14.28_1.0.0_ga, smarc-imx_3.10.53_1.1.0_ga, smarc-imx_3.14.28_1.0.0_ga, smarc-l5.0.0_1.0.0-ga

fs: block_dump missing dentry locking 

I think the block_dump output in __mark_inode_dirty is missing dentry locking.
Surely the i_dentry list can change any time, so we may not even *get* a
dentry there. If we do get one by chance, then it would appear to be able to
go away or get renamed at any time...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

Showing 1 changed file with 24 additions and 17 deletions Inline Diff

1 /* 1 /*
2 * fs/fs-writeback.c 2 * fs/fs-writeback.c
3 * 3 *
4 * Copyright (C) 2002, Linus Torvalds. 4 * Copyright (C) 2002, Linus Torvalds.
5 * 5 *
6 * Contains all the functions related to writing back and waiting 6 * Contains all the functions related to writing back and waiting
7 * upon dirty inodes against superblocks, and writing back dirty 7 * upon dirty inodes against superblocks, and writing back dirty
8 * pages against inodes. ie: data writeback. Writeout of the 8 * pages against inodes. ie: data writeback. Writeout of the
9 * inode itself is not handled here. 9 * inode itself is not handled here.
10 * 10 *
11 * 10Apr2002 Andrew Morton 11 * 10Apr2002 Andrew Morton
12 * Split out of fs/inode.c 12 * Split out of fs/inode.c
13 * Additions for address_space-based writeback 13 * Additions for address_space-based writeback
14 */ 14 */
15 15
16 #include <linux/kernel.h> 16 #include <linux/kernel.h>
17 #include <linux/module.h> 17 #include <linux/module.h>
18 #include <linux/spinlock.h> 18 #include <linux/spinlock.h>
19 #include <linux/sched.h> 19 #include <linux/sched.h>
20 #include <linux/fs.h> 20 #include <linux/fs.h>
21 #include <linux/mm.h> 21 #include <linux/mm.h>
22 #include <linux/writeback.h> 22 #include <linux/writeback.h>
23 #include <linux/blkdev.h> 23 #include <linux/blkdev.h>
24 #include <linux/backing-dev.h> 24 #include <linux/backing-dev.h>
25 #include <linux/buffer_head.h> 25 #include <linux/buffer_head.h>
26 #include "internal.h" 26 #include "internal.h"
27 27
28 28
29 /** 29 /**
30 * writeback_acquire - attempt to get exclusive writeback access to a device 30 * writeback_acquire - attempt to get exclusive writeback access to a device
31 * @bdi: the device's backing_dev_info structure 31 * @bdi: the device's backing_dev_info structure
32 * 32 *
33 * It is a waste of resources to have more than one pdflush thread blocked on 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 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. 35 * via a flag in the request_queue's backing_dev_info.state.
36 * 36 *
37 * Non-request_queue-backed address_spaces will share default_backing_dev_info, 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 38 * unless they implement their own. Which is somewhat inefficient, as this
39 * may prevent concurrent writeback against multiple devices. 39 * may prevent concurrent writeback against multiple devices.
40 */ 40 */
41 static int writeback_acquire(struct backing_dev_info *bdi) 41 static int writeback_acquire(struct backing_dev_info *bdi)
42 { 42 {
43 return !test_and_set_bit(BDI_pdflush, &bdi->state); 43 return !test_and_set_bit(BDI_pdflush, &bdi->state);
44 } 44 }
45 45
46 /** 46 /**
47 * writeback_in_progress - determine whether there is writeback in progress 47 * writeback_in_progress - determine whether there is writeback in progress
48 * @bdi: the device's backing_dev_info structure. 48 * @bdi: the device's backing_dev_info structure.
49 * 49 *
50 * Determine whether there is writeback in progress against a backing device. 50 * Determine whether there is writeback in progress against a backing device.
51 */ 51 */
52 int writeback_in_progress(struct backing_dev_info *bdi) 52 int writeback_in_progress(struct backing_dev_info *bdi)
53 { 53 {
54 return test_bit(BDI_pdflush, &bdi->state); 54 return test_bit(BDI_pdflush, &bdi->state);
55 } 55 }
56 56
57 /** 57 /**
58 * writeback_release - relinquish exclusive writeback access against a device. 58 * writeback_release - relinquish exclusive writeback access against a device.
59 * @bdi: the device's backing_dev_info structure 59 * @bdi: the device's backing_dev_info structure
60 */ 60 */
61 static void writeback_release(struct backing_dev_info *bdi) 61 static void writeback_release(struct backing_dev_info *bdi)
62 { 62 {
63 BUG_ON(!writeback_in_progress(bdi)); 63 BUG_ON(!writeback_in_progress(bdi));
64 clear_bit(BDI_pdflush, &bdi->state); 64 clear_bit(BDI_pdflush, &bdi->state);
65 } 65 }
66 66
67 static 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
67 /** 89 /**
68 * __mark_inode_dirty - internal function 90 * __mark_inode_dirty - internal function
69 * @inode: inode to mark 91 * @inode: inode to mark
70 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 92 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
71 * Mark an inode as dirty. Callers should use mark_inode_dirty or 93 * Mark an inode as dirty. Callers should use mark_inode_dirty or
72 * mark_inode_dirty_sync. 94 * mark_inode_dirty_sync.
73 * 95 *
74 * Put the inode on the super block's dirty list. 96 * Put the inode on the super block's dirty list.
75 * 97 *
76 * CAREFUL! We mark it dirty unconditionally, but move it onto the 98 * CAREFUL! We mark it dirty unconditionally, but move it onto the
77 * dirty list only if it is hashed or if it refers to a blockdev. 99 * dirty list only if it is hashed or if it refers to a blockdev.
78 * If it was not hashed, it will never be added to the dirty list 100 * If it was not hashed, it will never be added to the dirty list
79 * even if it is later hashed, as it will have been marked dirty already. 101 * even if it is later hashed, as it will have been marked dirty already.
80 * 102 *
81 * In short, make sure you hash any inodes _before_ you start marking 103 * In short, make sure you hash any inodes _before_ you start marking
82 * them dirty. 104 * them dirty.
83 * 105 *
84 * This function *must* be atomic for the I_DIRTY_PAGES case - 106 * This function *must* be atomic for the I_DIRTY_PAGES case -
85 * set_page_dirty() is called under spinlock in several places. 107 * set_page_dirty() is called under spinlock in several places.
86 * 108 *
87 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 109 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
88 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 110 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
89 * the kernel-internal blockdev inode represents the dirtying time of the 111 * the kernel-internal blockdev inode represents the dirtying time of the
90 * blockdev's pages. This is why for I_DIRTY_PAGES we always use 112 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
91 * page->mapping->host, so the page-dirtying time is recorded in the internal 113 * page->mapping->host, so the page-dirtying time is recorded in the internal
92 * blockdev inode. 114 * blockdev inode.
93 */ 115 */
94 void __mark_inode_dirty(struct inode *inode, int flags) 116 void __mark_inode_dirty(struct inode *inode, int flags)
95 { 117 {
96 struct super_block *sb = inode->i_sb; 118 struct super_block *sb = inode->i_sb;
97 119
98 /* 120 /*
99 * Don't do this for I_DIRTY_PAGES - that doesn't actually 121 * Don't do this for I_DIRTY_PAGES - that doesn't actually
100 * dirty the inode itself 122 * dirty the inode itself
101 */ 123 */
102 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 124 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
103 if (sb->s_op->dirty_inode) 125 if (sb->s_op->dirty_inode)
104 sb->s_op->dirty_inode(inode); 126 sb->s_op->dirty_inode(inode);
105 } 127 }
106 128
107 /* 129 /*
108 * make sure that changes are seen by all cpus before we test i_state 130 * make sure that changes are seen by all cpus before we test i_state
109 * -- mikulas 131 * -- mikulas
110 */ 132 */
111 smp_mb(); 133 smp_mb();
112 134
113 /* avoid the locking if we can */ 135 /* avoid the locking if we can */
114 if ((inode->i_state & flags) == flags) 136 if ((inode->i_state & flags) == flags)
115 return; 137 return;
116 138
117 if (unlikely(block_dump)) { 139 if (unlikely(block_dump))
118 struct dentry *dentry = NULL; 140 block_dump___mark_inode_dirty(inode);
119 const char *name = "?";
120
121 if (!list_empty(&inode->i_dentry)) {
122 dentry = list_entry(inode->i_dentry.next,
123 struct dentry, d_alias);
124 if (dentry && dentry->d_name.name)
125 name = (const char *) dentry->d_name.name;
126 }
127
128 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev"))
129 printk(KERN_DEBUG
130 "%s(%d): dirtied inode %lu (%s) on %s\n",
131 current->comm, task_pid_nr(current), inode->i_ino,
132 name, inode->i_sb->s_id);
133 }
134 141
135 spin_lock(&inode_lock); 142 spin_lock(&inode_lock);
136 if ((inode->i_state & flags) != flags) { 143 if ((inode->i_state & flags) != flags) {
137 const int was_dirty = inode->i_state & I_DIRTY; 144 const int was_dirty = inode->i_state & I_DIRTY;
138 145
139 inode->i_state |= flags; 146 inode->i_state |= flags;
140 147
141 /* 148 /*
142 * If the inode is being synced, just update its dirty state. 149 * If the inode is being synced, just update its dirty state.
143 * The unlocker will place the inode on the appropriate 150 * The unlocker will place the inode on the appropriate
144 * superblock list, based upon its state. 151 * superblock list, based upon its state.
145 */ 152 */
146 if (inode->i_state & I_SYNC) 153 if (inode->i_state & I_SYNC)
147 goto out; 154 goto out;
148 155
149 /* 156 /*
150 * Only add valid (hashed) inodes to the superblock's 157 * Only add valid (hashed) inodes to the superblock's
151 * dirty list. Add blockdev inodes as well. 158 * dirty list. Add blockdev inodes as well.
152 */ 159 */
153 if (!S_ISBLK(inode->i_mode)) { 160 if (!S_ISBLK(inode->i_mode)) {
154 if (hlist_unhashed(&inode->i_hash)) 161 if (hlist_unhashed(&inode->i_hash))
155 goto out; 162 goto out;
156 } 163 }
157 if (inode->i_state & (I_FREEING|I_CLEAR)) 164 if (inode->i_state & (I_FREEING|I_CLEAR))
158 goto out; 165 goto out;
159 166
160 /* 167 /*
161 * If the inode was already on s_dirty/s_io/s_more_io, don't 168 * If the inode was already on s_dirty/s_io/s_more_io, don't
162 * reposition it (that would break s_dirty time-ordering). 169 * reposition it (that would break s_dirty time-ordering).
163 */ 170 */
164 if (!was_dirty) { 171 if (!was_dirty) {
165 inode->dirtied_when = jiffies; 172 inode->dirtied_when = jiffies;
166 list_move(&inode->i_list, &sb->s_dirty); 173 list_move(&inode->i_list, &sb->s_dirty);
167 } 174 }
168 } 175 }
169 out: 176 out:
170 spin_unlock(&inode_lock); 177 spin_unlock(&inode_lock);
171 } 178 }
172 179
173 EXPORT_SYMBOL(__mark_inode_dirty); 180 EXPORT_SYMBOL(__mark_inode_dirty);
174 181
175 static int write_inode(struct inode *inode, int sync) 182 static int write_inode(struct inode *inode, int sync)
176 { 183 {
177 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) 184 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
178 return inode->i_sb->s_op->write_inode(inode, sync); 185 return inode->i_sb->s_op->write_inode(inode, sync);
179 return 0; 186 return 0;
180 } 187 }
181 188
182 /* 189 /*
183 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 190 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
184 * furthest end of its superblock's dirty-inode list. 191 * furthest end of its superblock's dirty-inode list.
185 * 192 *
186 * Before stamping the inode's ->dirtied_when, we check to see whether it is 193 * Before stamping the inode's ->dirtied_when, we check to see whether it is
187 * already the most-recently-dirtied inode on the s_dirty list. If that is 194 * already the most-recently-dirtied inode on the s_dirty list. If that is
188 * the case then the inode must have been redirtied while it was being written 195 * the case then the inode must have been redirtied while it was being written
189 * out and we don't reset its dirtied_when. 196 * out and we don't reset its dirtied_when.
190 */ 197 */
191 static void redirty_tail(struct inode *inode) 198 static void redirty_tail(struct inode *inode)
192 { 199 {
193 struct super_block *sb = inode->i_sb; 200 struct super_block *sb = inode->i_sb;
194 201
195 if (!list_empty(&sb->s_dirty)) { 202 if (!list_empty(&sb->s_dirty)) {
196 struct inode *tail_inode; 203 struct inode *tail_inode;
197 204
198 tail_inode = list_entry(sb->s_dirty.next, struct inode, i_list); 205 tail_inode = list_entry(sb->s_dirty.next, struct inode, i_list);
199 if (time_before(inode->dirtied_when, 206 if (time_before(inode->dirtied_when,
200 tail_inode->dirtied_when)) 207 tail_inode->dirtied_when))
201 inode->dirtied_when = jiffies; 208 inode->dirtied_when = jiffies;
202 } 209 }
203 list_move(&inode->i_list, &sb->s_dirty); 210 list_move(&inode->i_list, &sb->s_dirty);
204 } 211 }
205 212
206 /* 213 /*
207 * requeue inode for re-scanning after sb->s_io list is exhausted. 214 * requeue inode for re-scanning after sb->s_io list is exhausted.
208 */ 215 */
209 static void requeue_io(struct inode *inode) 216 static void requeue_io(struct inode *inode)
210 { 217 {
211 list_move(&inode->i_list, &inode->i_sb->s_more_io); 218 list_move(&inode->i_list, &inode->i_sb->s_more_io);
212 } 219 }
213 220
214 static void inode_sync_complete(struct inode *inode) 221 static void inode_sync_complete(struct inode *inode)
215 { 222 {
216 /* 223 /*
217 * Prevent speculative execution through spin_unlock(&inode_lock); 224 * Prevent speculative execution through spin_unlock(&inode_lock);
218 */ 225 */
219 smp_mb(); 226 smp_mb();
220 wake_up_bit(&inode->i_state, __I_SYNC); 227 wake_up_bit(&inode->i_state, __I_SYNC);
221 } 228 }
222 229
223 static bool inode_dirtied_after(struct inode *inode, unsigned long t) 230 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
224 { 231 {
225 bool ret = time_after(inode->dirtied_when, t); 232 bool ret = time_after(inode->dirtied_when, t);
226 #ifndef CONFIG_64BIT 233 #ifndef CONFIG_64BIT
227 /* 234 /*
228 * For inodes being constantly redirtied, dirtied_when can get stuck. 235 * For inodes being constantly redirtied, dirtied_when can get stuck.
229 * It _appears_ to be in the future, but is actually in distant past. 236 * It _appears_ to be in the future, but is actually in distant past.
230 * This test is necessary to prevent such wrapped-around relative times 237 * This test is necessary to prevent such wrapped-around relative times
231 * from permanently stopping the whole pdflush writeback. 238 * from permanently stopping the whole pdflush writeback.
232 */ 239 */
233 ret = ret && time_before_eq(inode->dirtied_when, jiffies); 240 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
234 #endif 241 #endif
235 return ret; 242 return ret;
236 } 243 }
237 244
238 /* 245 /*
239 * Move expired dirty inodes from @delaying_queue to @dispatch_queue. 246 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
240 */ 247 */
241 static void move_expired_inodes(struct list_head *delaying_queue, 248 static void move_expired_inodes(struct list_head *delaying_queue,
242 struct list_head *dispatch_queue, 249 struct list_head *dispatch_queue,
243 unsigned long *older_than_this) 250 unsigned long *older_than_this)
244 { 251 {
245 while (!list_empty(delaying_queue)) { 252 while (!list_empty(delaying_queue)) {
246 struct inode *inode = list_entry(delaying_queue->prev, 253 struct inode *inode = list_entry(delaying_queue->prev,
247 struct inode, i_list); 254 struct inode, i_list);
248 if (older_than_this && 255 if (older_than_this &&
249 inode_dirtied_after(inode, *older_than_this)) 256 inode_dirtied_after(inode, *older_than_this))
250 break; 257 break;
251 list_move(&inode->i_list, dispatch_queue); 258 list_move(&inode->i_list, dispatch_queue);
252 } 259 }
253 } 260 }
254 261
255 /* 262 /*
256 * Queue all expired dirty inodes for io, eldest first. 263 * Queue all expired dirty inodes for io, eldest first.
257 */ 264 */
258 static void queue_io(struct super_block *sb, 265 static void queue_io(struct super_block *sb,
259 unsigned long *older_than_this) 266 unsigned long *older_than_this)
260 { 267 {
261 list_splice_init(&sb->s_more_io, sb->s_io.prev); 268 list_splice_init(&sb->s_more_io, sb->s_io.prev);
262 move_expired_inodes(&sb->s_dirty, &sb->s_io, older_than_this); 269 move_expired_inodes(&sb->s_dirty, &sb->s_io, older_than_this);
263 } 270 }
264 271
265 int sb_has_dirty_inodes(struct super_block *sb) 272 int sb_has_dirty_inodes(struct super_block *sb)
266 { 273 {
267 return !list_empty(&sb->s_dirty) || 274 return !list_empty(&sb->s_dirty) ||
268 !list_empty(&sb->s_io) || 275 !list_empty(&sb->s_io) ||
269 !list_empty(&sb->s_more_io); 276 !list_empty(&sb->s_more_io);
270 } 277 }
271 EXPORT_SYMBOL(sb_has_dirty_inodes); 278 EXPORT_SYMBOL(sb_has_dirty_inodes);
272 279
273 /* 280 /*
274 * Write a single inode's dirty pages and inode data out to disk. 281 * Write a single inode's dirty pages and inode data out to disk.
275 * If `wait' is set, wait on the writeout. 282 * If `wait' is set, wait on the writeout.
276 * 283 *
277 * The whole writeout design is quite complex and fragile. We want to avoid 284 * The whole writeout design is quite complex and fragile. We want to avoid
278 * starvation of particular inodes when others are being redirtied, prevent 285 * starvation of particular inodes when others are being redirtied, prevent
279 * livelocks, etc. 286 * livelocks, etc.
280 * 287 *
281 * Called under inode_lock. 288 * Called under inode_lock.
282 */ 289 */
283 static int 290 static int
284 __sync_single_inode(struct inode *inode, struct writeback_control *wbc) 291 __sync_single_inode(struct inode *inode, struct writeback_control *wbc)
285 { 292 {
286 unsigned dirty; 293 unsigned dirty;
287 struct address_space *mapping = inode->i_mapping; 294 struct address_space *mapping = inode->i_mapping;
288 int wait = wbc->sync_mode == WB_SYNC_ALL; 295 int wait = wbc->sync_mode == WB_SYNC_ALL;
289 int ret; 296 int ret;
290 297
291 BUG_ON(inode->i_state & I_SYNC); 298 BUG_ON(inode->i_state & I_SYNC);
292 299
293 /* Set I_SYNC, reset I_DIRTY */ 300 /* Set I_SYNC, reset I_DIRTY */
294 dirty = inode->i_state & I_DIRTY; 301 dirty = inode->i_state & I_DIRTY;
295 inode->i_state |= I_SYNC; 302 inode->i_state |= I_SYNC;
296 inode->i_state &= ~I_DIRTY; 303 inode->i_state &= ~I_DIRTY;
297 304
298 spin_unlock(&inode_lock); 305 spin_unlock(&inode_lock);
299 306
300 ret = do_writepages(mapping, wbc); 307 ret = do_writepages(mapping, wbc);
301 308
302 /* Don't write the inode if only I_DIRTY_PAGES was set */ 309 /* Don't write the inode if only I_DIRTY_PAGES was set */
303 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 310 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
304 int err = write_inode(inode, wait); 311 int err = write_inode(inode, wait);
305 if (ret == 0) 312 if (ret == 0)
306 ret = err; 313 ret = err;
307 } 314 }
308 315
309 if (wait) { 316 if (wait) {
310 int err = filemap_fdatawait(mapping); 317 int err = filemap_fdatawait(mapping);
311 if (ret == 0) 318 if (ret == 0)
312 ret = err; 319 ret = err;
313 } 320 }
314 321
315 spin_lock(&inode_lock); 322 spin_lock(&inode_lock);
316 inode->i_state &= ~I_SYNC; 323 inode->i_state &= ~I_SYNC;
317 if (!(inode->i_state & I_FREEING)) { 324 if (!(inode->i_state & I_FREEING)) {
318 if (!(inode->i_state & I_DIRTY) && 325 if (!(inode->i_state & I_DIRTY) &&
319 mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { 326 mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
320 /* 327 /*
321 * We didn't write back all the pages. nfs_writepages() 328 * We didn't write back all the pages. nfs_writepages()
322 * sometimes bales out without doing anything. Redirty 329 * sometimes bales out without doing anything. Redirty
323 * the inode; Move it from s_io onto s_more_io/s_dirty. 330 * the inode; Move it from s_io onto s_more_io/s_dirty.
324 */ 331 */
325 /* 332 /*
326 * akpm: if the caller was the kupdate function we put 333 * akpm: if the caller was the kupdate function we put
327 * this inode at the head of s_dirty so it gets first 334 * this inode at the head of s_dirty so it gets first
328 * consideration. Otherwise, move it to the tail, for 335 * consideration. Otherwise, move it to the tail, for
329 * the reasons described there. I'm not really sure 336 * the reasons described there. I'm not really sure
330 * how much sense this makes. Presumably I had a good 337 * how much sense this makes. Presumably I had a good
331 * reasons for doing it this way, and I'd rather not 338 * reasons for doing it this way, and I'd rather not
332 * muck with it at present. 339 * muck with it at present.
333 */ 340 */
334 if (wbc->for_kupdate) { 341 if (wbc->for_kupdate) {
335 /* 342 /*
336 * For the kupdate function we move the inode 343 * For the kupdate function we move the inode
337 * to s_more_io so it will get more writeout as 344 * to s_more_io so it will get more writeout as
338 * soon as the queue becomes uncongested. 345 * soon as the queue becomes uncongested.
339 */ 346 */
340 inode->i_state |= I_DIRTY_PAGES; 347 inode->i_state |= I_DIRTY_PAGES;
341 if (wbc->nr_to_write <= 0) { 348 if (wbc->nr_to_write <= 0) {
342 /* 349 /*
343 * slice used up: queue for next turn 350 * slice used up: queue for next turn
344 */ 351 */
345 requeue_io(inode); 352 requeue_io(inode);
346 } else { 353 } else {
347 /* 354 /*
348 * somehow blocked: retry later 355 * somehow blocked: retry later
349 */ 356 */
350 redirty_tail(inode); 357 redirty_tail(inode);
351 } 358 }
352 } else { 359 } else {
353 /* 360 /*
354 * Otherwise fully redirty the inode so that 361 * Otherwise fully redirty the inode so that
355 * other inodes on this superblock will get some 362 * other inodes on this superblock will get some
356 * writeout. Otherwise heavy writing to one 363 * writeout. Otherwise heavy writing to one
357 * file would indefinitely suspend writeout of 364 * file would indefinitely suspend writeout of
358 * all the other files. 365 * all the other files.
359 */ 366 */
360 inode->i_state |= I_DIRTY_PAGES; 367 inode->i_state |= I_DIRTY_PAGES;
361 redirty_tail(inode); 368 redirty_tail(inode);
362 } 369 }
363 } else if (inode->i_state & I_DIRTY) { 370 } else if (inode->i_state & I_DIRTY) {
364 /* 371 /*
365 * Someone redirtied the inode while were writing back 372 * Someone redirtied the inode while were writing back
366 * the pages. 373 * the pages.
367 */ 374 */
368 redirty_tail(inode); 375 redirty_tail(inode);
369 } else if (atomic_read(&inode->i_count)) { 376 } else if (atomic_read(&inode->i_count)) {
370 /* 377 /*
371 * The inode is clean, inuse 378 * The inode is clean, inuse
372 */ 379 */
373 list_move(&inode->i_list, &inode_in_use); 380 list_move(&inode->i_list, &inode_in_use);
374 } else { 381 } else {
375 /* 382 /*
376 * The inode is clean, unused 383 * The inode is clean, unused
377 */ 384 */
378 list_move(&inode->i_list, &inode_unused); 385 list_move(&inode->i_list, &inode_unused);
379 } 386 }
380 } 387 }
381 inode_sync_complete(inode); 388 inode_sync_complete(inode);
382 return ret; 389 return ret;
383 } 390 }
384 391
385 /* 392 /*
386 * Write out an inode's dirty pages. Called under inode_lock. Either the 393 * Write out an inode's dirty pages. Called under inode_lock. Either the
387 * caller has ref on the inode (either via __iget or via syscall against an fd) 394 * caller has ref on the inode (either via __iget or via syscall against an fd)
388 * or the inode has I_WILL_FREE set (via generic_forget_inode) 395 * or the inode has I_WILL_FREE set (via generic_forget_inode)
389 */ 396 */
390 static int 397 static int
391 __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 398 __writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
392 { 399 {
393 wait_queue_head_t *wqh; 400 wait_queue_head_t *wqh;
394 401
395 if (!atomic_read(&inode->i_count)) 402 if (!atomic_read(&inode->i_count))
396 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 403 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
397 else 404 else
398 WARN_ON(inode->i_state & I_WILL_FREE); 405 WARN_ON(inode->i_state & I_WILL_FREE);
399 406
400 if ((wbc->sync_mode != WB_SYNC_ALL) && (inode->i_state & I_SYNC)) { 407 if ((wbc->sync_mode != WB_SYNC_ALL) && (inode->i_state & I_SYNC)) {
401 /* 408 /*
402 * We're skipping this inode because it's locked, and we're not 409 * We're skipping this inode because it's locked, and we're not
403 * doing writeback-for-data-integrity. Move it to s_more_io so 410 * doing writeback-for-data-integrity. Move it to s_more_io so
404 * that writeback can proceed with the other inodes on s_io. 411 * that writeback can proceed with the other inodes on s_io.
405 * We'll have another go at writing back this inode when we 412 * We'll have another go at writing back this inode when we
406 * completed a full scan of s_io. 413 * completed a full scan of s_io.
407 */ 414 */
408 requeue_io(inode); 415 requeue_io(inode);
409 return 0; 416 return 0;
410 } 417 }
411 418
412 /* 419 /*
413 * It's a data-integrity sync. We must wait. 420 * It's a data-integrity sync. We must wait.
414 */ 421 */
415 if (inode->i_state & I_SYNC) { 422 if (inode->i_state & I_SYNC) {
416 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 423 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
417 424
418 wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 425 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
419 do { 426 do {
420 spin_unlock(&inode_lock); 427 spin_unlock(&inode_lock);
421 __wait_on_bit(wqh, &wq, inode_wait, 428 __wait_on_bit(wqh, &wq, inode_wait,
422 TASK_UNINTERRUPTIBLE); 429 TASK_UNINTERRUPTIBLE);
423 spin_lock(&inode_lock); 430 spin_lock(&inode_lock);
424 } while (inode->i_state & I_SYNC); 431 } while (inode->i_state & I_SYNC);
425 } 432 }
426 return __sync_single_inode(inode, wbc); 433 return __sync_single_inode(inode, wbc);
427 } 434 }
428 435
429 /* 436 /*
430 * Write out a superblock's list of dirty inodes. A wait will be performed 437 * Write out a superblock's list of dirty inodes. A wait will be performed
431 * upon no inodes, all inodes or the final one, depending upon sync_mode. 438 * upon no inodes, all inodes or the final one, depending upon sync_mode.
432 * 439 *
433 * If older_than_this is non-NULL, then only write out inodes which 440 * If older_than_this is non-NULL, then only write out inodes which
434 * had their first dirtying at a time earlier than *older_than_this. 441 * had their first dirtying at a time earlier than *older_than_this.
435 * 442 *
436 * If we're a pdflush thread, then implement pdflush collision avoidance 443 * If we're a pdflush thread, then implement pdflush collision avoidance
437 * against the entire list. 444 * against the entire list.
438 * 445 *
439 * If `bdi' is non-zero then we're being asked to writeback a specific queue. 446 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
440 * This function assumes that the blockdev superblock's inodes are backed by 447 * This function assumes that the blockdev superblock's inodes are backed by
441 * a variety of queues, so all inodes are searched. For other superblocks, 448 * a variety of queues, so all inodes are searched. For other superblocks,
442 * assume that all inodes are backed by the same queue. 449 * assume that all inodes are backed by the same queue.
443 * 450 *
444 * FIXME: this linear search could get expensive with many fileystems. But 451 * FIXME: this linear search could get expensive with many fileystems. But
445 * how to fix? We need to go from an address_space to all inodes which share 452 * how to fix? We need to go from an address_space to all inodes which share
446 * a queue with that address_space. (Easy: have a global "dirty superblocks" 453 * a queue with that address_space. (Easy: have a global "dirty superblocks"
447 * list). 454 * list).
448 * 455 *
449 * The inodes to be written are parked on sb->s_io. They are moved back onto 456 * The inodes to be written are parked on sb->s_io. They are moved back onto
450 * sb->s_dirty as they are selected for writing. This way, none can be missed 457 * sb->s_dirty as they are selected for writing. This way, none can be missed
451 * on the writer throttling path, and we get decent balancing between many 458 * on the writer throttling path, and we get decent balancing between many
452 * throttled threads: we don't want them all piling up on inode_sync_wait. 459 * throttled threads: we don't want them all piling up on inode_sync_wait.
453 */ 460 */
454 void generic_sync_sb_inodes(struct super_block *sb, 461 void generic_sync_sb_inodes(struct super_block *sb,
455 struct writeback_control *wbc) 462 struct writeback_control *wbc)
456 { 463 {
457 const unsigned long start = jiffies; /* livelock avoidance */ 464 const unsigned long start = jiffies; /* livelock avoidance */
458 int sync = wbc->sync_mode == WB_SYNC_ALL; 465 int sync = wbc->sync_mode == WB_SYNC_ALL;
459 466
460 spin_lock(&inode_lock); 467 spin_lock(&inode_lock);
461 if (!wbc->for_kupdate || list_empty(&sb->s_io)) 468 if (!wbc->for_kupdate || list_empty(&sb->s_io))
462 queue_io(sb, wbc->older_than_this); 469 queue_io(sb, wbc->older_than_this);
463 470
464 while (!list_empty(&sb->s_io)) { 471 while (!list_empty(&sb->s_io)) {
465 struct inode *inode = list_entry(sb->s_io.prev, 472 struct inode *inode = list_entry(sb->s_io.prev,
466 struct inode, i_list); 473 struct inode, i_list);
467 struct address_space *mapping = inode->i_mapping; 474 struct address_space *mapping = inode->i_mapping;
468 struct backing_dev_info *bdi = mapping->backing_dev_info; 475 struct backing_dev_info *bdi = mapping->backing_dev_info;
469 long pages_skipped; 476 long pages_skipped;
470 477
471 if (!bdi_cap_writeback_dirty(bdi)) { 478 if (!bdi_cap_writeback_dirty(bdi)) {
472 redirty_tail(inode); 479 redirty_tail(inode);
473 if (sb_is_blkdev_sb(sb)) { 480 if (sb_is_blkdev_sb(sb)) {
474 /* 481 /*
475 * Dirty memory-backed blockdev: the ramdisk 482 * Dirty memory-backed blockdev: the ramdisk
476 * driver does this. Skip just this inode 483 * driver does this. Skip just this inode
477 */ 484 */
478 continue; 485 continue;
479 } 486 }
480 /* 487 /*
481 * Dirty memory-backed inode against a filesystem other 488 * Dirty memory-backed inode against a filesystem other
482 * than the kernel-internal bdev filesystem. Skip the 489 * than the kernel-internal bdev filesystem. Skip the
483 * entire superblock. 490 * entire superblock.
484 */ 491 */
485 break; 492 break;
486 } 493 }
487 494
488 if (inode->i_state & I_NEW) { 495 if (inode->i_state & I_NEW) {
489 requeue_io(inode); 496 requeue_io(inode);
490 continue; 497 continue;
491 } 498 }
492 499
493 if (wbc->nonblocking && bdi_write_congested(bdi)) { 500 if (wbc->nonblocking && bdi_write_congested(bdi)) {
494 wbc->encountered_congestion = 1; 501 wbc->encountered_congestion = 1;
495 if (!sb_is_blkdev_sb(sb)) 502 if (!sb_is_blkdev_sb(sb))
496 break; /* Skip a congested fs */ 503 break; /* Skip a congested fs */
497 requeue_io(inode); 504 requeue_io(inode);
498 continue; /* Skip a congested blockdev */ 505 continue; /* Skip a congested blockdev */
499 } 506 }
500 507
501 if (wbc->bdi && bdi != wbc->bdi) { 508 if (wbc->bdi && bdi != wbc->bdi) {
502 if (!sb_is_blkdev_sb(sb)) 509 if (!sb_is_blkdev_sb(sb))
503 break; /* fs has the wrong queue */ 510 break; /* fs has the wrong queue */
504 requeue_io(inode); 511 requeue_io(inode);
505 continue; /* blockdev has wrong queue */ 512 continue; /* blockdev has wrong queue */
506 } 513 }
507 514
508 /* 515 /*
509 * Was this inode dirtied after sync_sb_inodes was called? 516 * Was this inode dirtied after sync_sb_inodes was called?
510 * This keeps sync from extra jobs and livelock. 517 * This keeps sync from extra jobs and livelock.
511 */ 518 */
512 if (inode_dirtied_after(inode, start)) 519 if (inode_dirtied_after(inode, start))
513 break; 520 break;
514 521
515 /* Is another pdflush already flushing this queue? */ 522 /* Is another pdflush already flushing this queue? */
516 if (current_is_pdflush() && !writeback_acquire(bdi)) 523 if (current_is_pdflush() && !writeback_acquire(bdi))
517 break; 524 break;
518 525
519 BUG_ON(inode->i_state & I_FREEING); 526 BUG_ON(inode->i_state & I_FREEING);
520 __iget(inode); 527 __iget(inode);
521 pages_skipped = wbc->pages_skipped; 528 pages_skipped = wbc->pages_skipped;
522 __writeback_single_inode(inode, wbc); 529 __writeback_single_inode(inode, wbc);
523 if (current_is_pdflush()) 530 if (current_is_pdflush())
524 writeback_release(bdi); 531 writeback_release(bdi);
525 if (wbc->pages_skipped != pages_skipped) { 532 if (wbc->pages_skipped != pages_skipped) {
526 /* 533 /*
527 * writeback is not making progress due to locked 534 * writeback is not making progress due to locked
528 * buffers. Skip this inode for now. 535 * buffers. Skip this inode for now.
529 */ 536 */
530 redirty_tail(inode); 537 redirty_tail(inode);
531 } 538 }
532 spin_unlock(&inode_lock); 539 spin_unlock(&inode_lock);
533 iput(inode); 540 iput(inode);
534 cond_resched(); 541 cond_resched();
535 spin_lock(&inode_lock); 542 spin_lock(&inode_lock);
536 if (wbc->nr_to_write <= 0) { 543 if (wbc->nr_to_write <= 0) {
537 wbc->more_io = 1; 544 wbc->more_io = 1;
538 break; 545 break;
539 } 546 }
540 if (!list_empty(&sb->s_more_io)) 547 if (!list_empty(&sb->s_more_io))
541 wbc->more_io = 1; 548 wbc->more_io = 1;
542 } 549 }
543 550
544 if (sync) { 551 if (sync) {
545 struct inode *inode, *old_inode = NULL; 552 struct inode *inode, *old_inode = NULL;
546 553
547 /* 554 /*
548 * Data integrity sync. Must wait for all pages under writeback, 555 * Data integrity sync. Must wait for all pages under writeback,
549 * because there may have been pages dirtied before our sync 556 * because there may have been pages dirtied before our sync
550 * call, but which had writeout started before we write it out. 557 * call, but which had writeout started before we write it out.
551 * In which case, the inode may not be on the dirty list, but 558 * In which case, the inode may not be on the dirty list, but
552 * we still have to wait for that writeout. 559 * we still have to wait for that writeout.
553 */ 560 */
554 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 561 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
555 struct address_space *mapping; 562 struct address_space *mapping;
556 563
557 if (inode->i_state & 564 if (inode->i_state &
558 (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW)) 565 (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW))
559 continue; 566 continue;
560 mapping = inode->i_mapping; 567 mapping = inode->i_mapping;
561 if (mapping->nrpages == 0) 568 if (mapping->nrpages == 0)
562 continue; 569 continue;
563 __iget(inode); 570 __iget(inode);
564 spin_unlock(&inode_lock); 571 spin_unlock(&inode_lock);
565 /* 572 /*
566 * We hold a reference to 'inode' so it couldn't have 573 * We hold a reference to 'inode' so it couldn't have
567 * been removed from s_inodes list while we dropped the 574 * been removed from s_inodes list while we dropped the
568 * inode_lock. We cannot iput the inode now as we can 575 * inode_lock. We cannot iput the inode now as we can
569 * be holding the last reference and we cannot iput it 576 * be holding the last reference and we cannot iput it
570 * under inode_lock. So we keep the reference and iput 577 * under inode_lock. So we keep the reference and iput
571 * it later. 578 * it later.
572 */ 579 */
573 iput(old_inode); 580 iput(old_inode);
574 old_inode = inode; 581 old_inode = inode;
575 582
576 filemap_fdatawait(mapping); 583 filemap_fdatawait(mapping);
577 584
578 cond_resched(); 585 cond_resched();
579 586
580 spin_lock(&inode_lock); 587 spin_lock(&inode_lock);
581 } 588 }
582 spin_unlock(&inode_lock); 589 spin_unlock(&inode_lock);
583 iput(old_inode); 590 iput(old_inode);
584 } else 591 } else
585 spin_unlock(&inode_lock); 592 spin_unlock(&inode_lock);
586 593
587 return; /* Leave any unwritten inodes on s_io */ 594 return; /* Leave any unwritten inodes on s_io */
588 } 595 }
589 EXPORT_SYMBOL_GPL(generic_sync_sb_inodes); 596 EXPORT_SYMBOL_GPL(generic_sync_sb_inodes);
590 597
591 static void sync_sb_inodes(struct super_block *sb, 598 static void sync_sb_inodes(struct super_block *sb,
592 struct writeback_control *wbc) 599 struct writeback_control *wbc)
593 { 600 {
594 generic_sync_sb_inodes(sb, wbc); 601 generic_sync_sb_inodes(sb, wbc);
595 } 602 }
596 603
597 /* 604 /*
598 * Start writeback of dirty pagecache data against all unlocked inodes. 605 * Start writeback of dirty pagecache data against all unlocked inodes.
599 * 606 *
600 * Note: 607 * Note:
601 * We don't need to grab a reference to superblock here. If it has non-empty 608 * We don't need to grab a reference to superblock here. If it has non-empty
602 * ->s_dirty it's hadn't been killed yet and kill_super() won't proceed 609 * ->s_dirty it's hadn't been killed yet and kill_super() won't proceed
603 * past sync_inodes_sb() until the ->s_dirty/s_io/s_more_io lists are all 610 * past sync_inodes_sb() until the ->s_dirty/s_io/s_more_io lists are all
604 * empty. Since __sync_single_inode() regains inode_lock before it finally moves 611 * empty. Since __sync_single_inode() regains inode_lock before it finally moves
605 * inode from superblock lists we are OK. 612 * inode from superblock lists we are OK.
606 * 613 *
607 * If `older_than_this' is non-zero then only flush inodes which have a 614 * If `older_than_this' is non-zero then only flush inodes which have a
608 * flushtime older than *older_than_this. 615 * flushtime older than *older_than_this.
609 * 616 *
610 * If `bdi' is non-zero then we will scan the first inode against each 617 * If `bdi' is non-zero then we will scan the first inode against each
611 * superblock until we find the matching ones. One group will be the dirty 618 * superblock until we find the matching ones. One group will be the dirty
612 * inodes against a filesystem. Then when we hit the dummy blockdev superblock, 619 * inodes against a filesystem. Then when we hit the dummy blockdev superblock,
613 * sync_sb_inodes will seekout the blockdev which matches `bdi'. Maybe not 620 * sync_sb_inodes will seekout the blockdev which matches `bdi'. Maybe not
614 * super-efficient but we're about to do a ton of I/O... 621 * super-efficient but we're about to do a ton of I/O...
615 */ 622 */
616 void 623 void
617 writeback_inodes(struct writeback_control *wbc) 624 writeback_inodes(struct writeback_control *wbc)
618 { 625 {
619 struct super_block *sb; 626 struct super_block *sb;
620 627
621 might_sleep(); 628 might_sleep();
622 spin_lock(&sb_lock); 629 spin_lock(&sb_lock);
623 restart: 630 restart:
624 list_for_each_entry_reverse(sb, &super_blocks, s_list) { 631 list_for_each_entry_reverse(sb, &super_blocks, s_list) {
625 if (sb_has_dirty_inodes(sb)) { 632 if (sb_has_dirty_inodes(sb)) {
626 /* we're making our own get_super here */ 633 /* we're making our own get_super here */
627 sb->s_count++; 634 sb->s_count++;
628 spin_unlock(&sb_lock); 635 spin_unlock(&sb_lock);
629 /* 636 /*
630 * If we can't get the readlock, there's no sense in 637 * If we can't get the readlock, there's no sense in
631 * waiting around, most of the time the FS is going to 638 * waiting around, most of the time the FS is going to
632 * be unmounted by the time it is released. 639 * be unmounted by the time it is released.
633 */ 640 */
634 if (down_read_trylock(&sb->s_umount)) { 641 if (down_read_trylock(&sb->s_umount)) {
635 if (sb->s_root) 642 if (sb->s_root)
636 sync_sb_inodes(sb, wbc); 643 sync_sb_inodes(sb, wbc);
637 up_read(&sb->s_umount); 644 up_read(&sb->s_umount);
638 } 645 }
639 spin_lock(&sb_lock); 646 spin_lock(&sb_lock);
640 if (__put_super_and_need_restart(sb)) 647 if (__put_super_and_need_restart(sb))
641 goto restart; 648 goto restart;
642 } 649 }
643 if (wbc->nr_to_write <= 0) 650 if (wbc->nr_to_write <= 0)
644 break; 651 break;
645 } 652 }
646 spin_unlock(&sb_lock); 653 spin_unlock(&sb_lock);
647 } 654 }
648 655
649 /* 656 /*
650 * writeback and wait upon the filesystem's dirty inodes. The caller will 657 * writeback and wait upon the filesystem's dirty inodes. The caller will
651 * do this in two passes - one to write, and one to wait. 658 * do this in two passes - one to write, and one to wait.
652 * 659 *
653 * A finite limit is set on the number of pages which will be written. 660 * A finite limit is set on the number of pages which will be written.
654 * To prevent infinite livelock of sys_sync(). 661 * To prevent infinite livelock of sys_sync().
655 * 662 *
656 * We add in the number of potentially dirty inodes, because each inode write 663 * We add in the number of potentially dirty inodes, because each inode write
657 * can dirty pagecache in the underlying blockdev. 664 * can dirty pagecache in the underlying blockdev.
658 */ 665 */
659 void sync_inodes_sb(struct super_block *sb, int wait) 666 void sync_inodes_sb(struct super_block *sb, int wait)
660 { 667 {
661 struct writeback_control wbc = { 668 struct writeback_control wbc = {
662 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 669 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
663 .range_start = 0, 670 .range_start = 0,
664 .range_end = LLONG_MAX, 671 .range_end = LLONG_MAX,
665 }; 672 };
666 673
667 if (!wait) { 674 if (!wait) {
668 unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); 675 unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY);
669 unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); 676 unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS);
670 677
671 wbc.nr_to_write = nr_dirty + nr_unstable + 678 wbc.nr_to_write = nr_dirty + nr_unstable +
672 (inodes_stat.nr_inodes - inodes_stat.nr_unused); 679 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
673 } else 680 } else
674 wbc.nr_to_write = LONG_MAX; /* doesn't actually matter */ 681 wbc.nr_to_write = LONG_MAX; /* doesn't actually matter */
675 682
676 sync_sb_inodes(sb, &wbc); 683 sync_sb_inodes(sb, &wbc);
677 } 684 }
678 685
679 /** 686 /**
680 * write_inode_now - write an inode to disk 687 * write_inode_now - write an inode to disk
681 * @inode: inode to write to disk 688 * @inode: inode to write to disk
682 * @sync: whether the write should be synchronous or not 689 * @sync: whether the write should be synchronous or not
683 * 690 *
684 * This function commits an inode to disk immediately if it is dirty. This is 691 * This function commits an inode to disk immediately if it is dirty. This is
685 * primarily needed by knfsd. 692 * primarily needed by knfsd.
686 * 693 *
687 * The caller must either have a ref on the inode or must have set I_WILL_FREE. 694 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
688 */ 695 */
689 int write_inode_now(struct inode *inode, int sync) 696 int write_inode_now(struct inode *inode, int sync)
690 { 697 {
691 int ret; 698 int ret;
692 struct writeback_control wbc = { 699 struct writeback_control wbc = {
693 .nr_to_write = LONG_MAX, 700 .nr_to_write = LONG_MAX,
694 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 701 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
695 .range_start = 0, 702 .range_start = 0,
696 .range_end = LLONG_MAX, 703 .range_end = LLONG_MAX,
697 }; 704 };
698 705
699 if (!mapping_cap_writeback_dirty(inode->i_mapping)) 706 if (!mapping_cap_writeback_dirty(inode->i_mapping))
700 wbc.nr_to_write = 0; 707 wbc.nr_to_write = 0;
701 708
702 might_sleep(); 709 might_sleep();
703 spin_lock(&inode_lock); 710 spin_lock(&inode_lock);
704 ret = __writeback_single_inode(inode, &wbc); 711 ret = __writeback_single_inode(inode, &wbc);
705 spin_unlock(&inode_lock); 712 spin_unlock(&inode_lock);
706 if (sync) 713 if (sync)
707 inode_sync_wait(inode); 714 inode_sync_wait(inode);
708 return ret; 715 return ret;
709 } 716 }
710 EXPORT_SYMBOL(write_inode_now); 717 EXPORT_SYMBOL(write_inode_now);
711 718
712 /** 719 /**
713 * sync_inode - write an inode and its pages to disk. 720 * sync_inode - write an inode and its pages to disk.
714 * @inode: the inode to sync 721 * @inode: the inode to sync
715 * @wbc: controls the writeback mode 722 * @wbc: controls the writeback mode
716 * 723 *
717 * sync_inode() will write an inode and its pages to disk. It will also 724 * sync_inode() will write an inode and its pages to disk. It will also
718 * correctly update the inode on its superblock's dirty inode lists and will 725 * correctly update the inode on its superblock's dirty inode lists and will
719 * update inode->i_state. 726 * update inode->i_state.
720 * 727 *
721 * The caller must have a ref on the inode. 728 * The caller must have a ref on the inode.
722 */ 729 */
723 int sync_inode(struct inode *inode, struct writeback_control *wbc) 730 int sync_inode(struct inode *inode, struct writeback_control *wbc)
724 { 731 {
725 int ret; 732 int ret;
726 733
727 spin_lock(&inode_lock); 734 spin_lock(&inode_lock);
728 ret = __writeback_single_inode(inode, wbc); 735 ret = __writeback_single_inode(inode, wbc);
729 spin_unlock(&inode_lock); 736 spin_unlock(&inode_lock);
730 return ret; 737 return ret;
731 } 738 }
732 EXPORT_SYMBOL(sync_inode); 739 EXPORT_SYMBOL(sync_inode);
733 740
734 /** 741 /**
735 * generic_osync_inode - flush all dirty data for a given inode to disk 742 * generic_osync_inode - flush all dirty data for a given inode to disk
736 * @inode: inode to write 743 * @inode: inode to write
737 * @mapping: the address_space that should be flushed 744 * @mapping: the address_space that should be flushed
738 * @what: what to write and wait upon 745 * @what: what to write and wait upon
739 * 746 *
740 * This can be called by file_write functions for files which have the 747 * This can be called by file_write functions for files which have the
741 * O_SYNC flag set, to flush dirty writes to disk. 748 * O_SYNC flag set, to flush dirty writes to disk.
742 * 749 *
743 * @what is a bitmask, specifying which part of the inode's data should be 750 * @what is a bitmask, specifying which part of the inode's data should be
744 * written and waited upon. 751 * written and waited upon.
745 * 752 *
746 * OSYNC_DATA: i_mapping's dirty data 753 * OSYNC_DATA: i_mapping's dirty data
747 * OSYNC_METADATA: the buffers at i_mapping->private_list 754 * OSYNC_METADATA: the buffers at i_mapping->private_list
748 * OSYNC_INODE: the inode itself 755 * OSYNC_INODE: the inode itself
749 */ 756 */
750 757
751 int generic_osync_inode(struct inode *inode, struct address_space *mapping, int what) 758 int generic_osync_inode(struct inode *inode, struct address_space *mapping, int what)
752 { 759 {
753 int err = 0; 760 int err = 0;
754 int need_write_inode_now = 0; 761 int need_write_inode_now = 0;
755 int err2; 762 int err2;
756 763
757 if (what & OSYNC_DATA) 764 if (what & OSYNC_DATA)
758 err = filemap_fdatawrite(mapping); 765 err = filemap_fdatawrite(mapping);
759 if (what & (OSYNC_METADATA|OSYNC_DATA)) { 766 if (what & (OSYNC_METADATA|OSYNC_DATA)) {
760 err2 = sync_mapping_buffers(mapping); 767 err2 = sync_mapping_buffers(mapping);
761 if (!err) 768 if (!err)
762 err = err2; 769 err = err2;
763 } 770 }
764 if (what & OSYNC_DATA) { 771 if (what & OSYNC_DATA) {
765 err2 = filemap_fdatawait(mapping); 772 err2 = filemap_fdatawait(mapping);
766 if (!err) 773 if (!err)
767 err = err2; 774 err = err2;
768 } 775 }
769 776
770 spin_lock(&inode_lock); 777 spin_lock(&inode_lock);
771 if ((inode->i_state & I_DIRTY) && 778 if ((inode->i_state & I_DIRTY) &&
772 ((what & OSYNC_INODE) || (inode->i_state & I_DIRTY_DATASYNC))) 779 ((what & OSYNC_INODE) || (inode->i_state & I_DIRTY_DATASYNC)))