Doug / smarc-fsl-linux-kernel

Download as
Browse Code ยป

Commit ae1b7f7d4b9ea587fda95c38301f4e72e8146634

Authored by Wu Fengguang
Committed by Jens Axboe
1 parent 7fbdea3232
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

writeback: cleanup writeback_single_inode() 

Make the if-else straight in writeback_single_inode().
No behavior change.

Cc: Jan Kara <jack@suse.cz>
Cc: Michael Rubin <mrubin@google.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

Showing 1 changed file with 7 additions and 8 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/kthread.h> 22 #include <linux/kthread.h>
23 #include <linux/freezer.h> 23 #include <linux/freezer.h>
24 #include <linux/writeback.h> 24 #include <linux/writeback.h>
25 #include <linux/blkdev.h> 25 #include <linux/blkdev.h>
26 #include <linux/backing-dev.h> 26 #include <linux/backing-dev.h>
27 #include <linux/buffer_head.h> 27 #include <linux/buffer_head.h>
28 #include "internal.h" 28 #include "internal.h"
29 29
30 #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info) 30 #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info)
31 31
32 /* 32 /*
33 * We don't actually have pdflush, but this one is exported though /proc... 33 * We don't actually have pdflush, but this one is exported though /proc...
34 */ 34 */
35 int nr_pdflush_threads; 35 int nr_pdflush_threads;
36 36
37 /* 37 /*
38 * Passed into wb_writeback(), essentially a subset of writeback_control 38 * Passed into wb_writeback(), essentially a subset of writeback_control
39 */ 39 */
40 struct wb_writeback_args { 40 struct wb_writeback_args {
41 long nr_pages; 41 long nr_pages;
42 struct super_block *sb; 42 struct super_block *sb;
43 enum writeback_sync_modes sync_mode; 43 enum writeback_sync_modes sync_mode;
44 int for_kupdate:1; 44 int for_kupdate:1;
45 int range_cyclic:1; 45 int range_cyclic:1;
46 int for_background:1; 46 int for_background:1;
47 }; 47 };
48 48
49 /* 49 /*
50 * Work items for the bdi_writeback threads 50 * Work items for the bdi_writeback threads
51 */ 51 */
52 struct bdi_work { 52 struct bdi_work {
53 struct list_head list; /* pending work list */ 53 struct list_head list; /* pending work list */
54 struct rcu_head rcu_head; /* for RCU free/clear of work */ 54 struct rcu_head rcu_head; /* for RCU free/clear of work */
55 55
56 unsigned long seen; /* threads that have seen this work */ 56 unsigned long seen; /* threads that have seen this work */
57 atomic_t pending; /* number of threads still to do work */ 57 atomic_t pending; /* number of threads still to do work */
58 58
59 struct wb_writeback_args args; /* writeback arguments */ 59 struct wb_writeback_args args; /* writeback arguments */
60 60
61 unsigned long state; /* flag bits, see WS_* */ 61 unsigned long state; /* flag bits, see WS_* */
62 }; 62 };
63 63
64 enum { 64 enum {
65 WS_USED_B = 0, 65 WS_USED_B = 0,
66 WS_ONSTACK_B, 66 WS_ONSTACK_B,
67 }; 67 };
68 68
69 #define WS_USED (1 << WS_USED_B) 69 #define WS_USED (1 << WS_USED_B)
70 #define WS_ONSTACK (1 << WS_ONSTACK_B) 70 #define WS_ONSTACK (1 << WS_ONSTACK_B)
71 71
72 static inline bool bdi_work_on_stack(struct bdi_work *work) 72 static inline bool bdi_work_on_stack(struct bdi_work *work)
73 { 73 {
74 return test_bit(WS_ONSTACK_B, &work->state); 74 return test_bit(WS_ONSTACK_B, &work->state);
75 } 75 }
76 76
77 static inline void bdi_work_init(struct bdi_work *work, 77 static inline void bdi_work_init(struct bdi_work *work,
78 struct wb_writeback_args *args) 78 struct wb_writeback_args *args)
79 { 79 {
80 INIT_RCU_HEAD(&work->rcu_head); 80 INIT_RCU_HEAD(&work->rcu_head);
81 work->args = *args; 81 work->args = *args;
82 work->state = WS_USED; 82 work->state = WS_USED;
83 } 83 }
84 84
85 /** 85 /**
86 * writeback_in_progress - determine whether there is writeback in progress 86 * writeback_in_progress - determine whether there is writeback in progress
87 * @bdi: the device's backing_dev_info structure. 87 * @bdi: the device's backing_dev_info structure.
88 * 88 *
89 * Determine whether there is writeback waiting to be handled against a 89 * Determine whether there is writeback waiting to be handled against a
90 * backing device. 90 * backing device.
91 */ 91 */
92 int writeback_in_progress(struct backing_dev_info *bdi) 92 int writeback_in_progress(struct backing_dev_info *bdi)
93 { 93 {
94 return !list_empty(&bdi->work_list); 94 return !list_empty(&bdi->work_list);
95 } 95 }
96 96
97 static void bdi_work_clear(struct bdi_work *work) 97 static void bdi_work_clear(struct bdi_work *work)
98 { 98 {
99 clear_bit(WS_USED_B, &work->state); 99 clear_bit(WS_USED_B, &work->state);
100 smp_mb__after_clear_bit(); 100 smp_mb__after_clear_bit();
101 /* 101 /*
102 * work can have disappeared at this point. bit waitq functions 102 * work can have disappeared at this point. bit waitq functions
103 * should be able to tolerate this, provided bdi_sched_wait does 103 * should be able to tolerate this, provided bdi_sched_wait does
104 * not dereference it's pointer argument. 104 * not dereference it's pointer argument.
105 */ 105 */
106 wake_up_bit(&work->state, WS_USED_B); 106 wake_up_bit(&work->state, WS_USED_B);
107 } 107 }
108 108
109 static void bdi_work_free(struct rcu_head *head) 109 static void bdi_work_free(struct rcu_head *head)
110 { 110 {
111 struct bdi_work *work = container_of(head, struct bdi_work, rcu_head); 111 struct bdi_work *work = container_of(head, struct bdi_work, rcu_head);
112 112
113 if (!bdi_work_on_stack(work)) 113 if (!bdi_work_on_stack(work))
114 kfree(work); 114 kfree(work);
115 else 115 else
116 bdi_work_clear(work); 116 bdi_work_clear(work);
117 } 117 }
118 118
119 static void wb_work_complete(struct bdi_work *work) 119 static void wb_work_complete(struct bdi_work *work)
120 { 120 {
121 const enum writeback_sync_modes sync_mode = work->args.sync_mode; 121 const enum writeback_sync_modes sync_mode = work->args.sync_mode;
122 int onstack = bdi_work_on_stack(work); 122 int onstack = bdi_work_on_stack(work);
123 123
124 /* 124 /*
125 * For allocated work, we can clear the done/seen bit right here. 125 * For allocated work, we can clear the done/seen bit right here.
126 * For on-stack work, we need to postpone both the clear and free 126 * For on-stack work, we need to postpone both the clear and free
127 * to after the RCU grace period, since the stack could be invalidated 127 * to after the RCU grace period, since the stack could be invalidated
128 * as soon as bdi_work_clear() has done the wakeup. 128 * as soon as bdi_work_clear() has done the wakeup.
129 */ 129 */
130 if (!onstack) 130 if (!onstack)
131 bdi_work_clear(work); 131 bdi_work_clear(work);
132 if (sync_mode == WB_SYNC_NONE || onstack) 132 if (sync_mode == WB_SYNC_NONE || onstack)
133 call_rcu(&work->rcu_head, bdi_work_free); 133 call_rcu(&work->rcu_head, bdi_work_free);
134 } 134 }
135 135
136 static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work) 136 static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work)
137 { 137 {
138 /* 138 /*
139 * The caller has retrieved the work arguments from this work, 139 * The caller has retrieved the work arguments from this work,
140 * drop our reference. If this is the last ref, delete and free it 140 * drop our reference. If this is the last ref, delete and free it
141 */ 141 */
142 if (atomic_dec_and_test(&work->pending)) { 142 if (atomic_dec_and_test(&work->pending)) {
143 struct backing_dev_info *bdi = wb->bdi; 143 struct backing_dev_info *bdi = wb->bdi;
144 144
145 spin_lock(&bdi->wb_lock); 145 spin_lock(&bdi->wb_lock);
146 list_del_rcu(&work->list); 146 list_del_rcu(&work->list);
147 spin_unlock(&bdi->wb_lock); 147 spin_unlock(&bdi->wb_lock);
148 148
149 wb_work_complete(work); 149 wb_work_complete(work);
150 } 150 }
151 } 151 }
152 152
153 static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work) 153 static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work)
154 { 154 {
155 work->seen = bdi->wb_mask; 155 work->seen = bdi->wb_mask;
156 BUG_ON(!work->seen); 156 BUG_ON(!work->seen);
157 atomic_set(&work->pending, bdi->wb_cnt); 157 atomic_set(&work->pending, bdi->wb_cnt);
158 BUG_ON(!bdi->wb_cnt); 158 BUG_ON(!bdi->wb_cnt);
159 159
160 /* 160 /*
161 * list_add_tail_rcu() contains the necessary barriers to 161 * list_add_tail_rcu() contains the necessary barriers to
162 * make sure the above stores are seen before the item is 162 * make sure the above stores are seen before the item is
163 * noticed on the list 163 * noticed on the list
164 */ 164 */
165 spin_lock(&bdi->wb_lock); 165 spin_lock(&bdi->wb_lock);
166 list_add_tail_rcu(&work->list, &bdi->work_list); 166 list_add_tail_rcu(&work->list, &bdi->work_list);
167 spin_unlock(&bdi->wb_lock); 167 spin_unlock(&bdi->wb_lock);
168 168
169 /* 169 /*
170 * If the default thread isn't there, make sure we add it. When 170 * If the default thread isn't there, make sure we add it. When
171 * it gets created and wakes up, we'll run this work. 171 * it gets created and wakes up, we'll run this work.
172 */ 172 */
173 if (unlikely(list_empty_careful(&bdi->wb_list))) 173 if (unlikely(list_empty_careful(&bdi->wb_list)))
174 wake_up_process(default_backing_dev_info.wb.task); 174 wake_up_process(default_backing_dev_info.wb.task);
175 else { 175 else {
176 struct bdi_writeback *wb = &bdi->wb; 176 struct bdi_writeback *wb = &bdi->wb;
177 177
178 if (wb->task) 178 if (wb->task)
179 wake_up_process(wb->task); 179 wake_up_process(wb->task);
180 } 180 }
181 } 181 }
182 182
183 /* 183 /*
184 * Used for on-stack allocated work items. The caller needs to wait until 184 * Used for on-stack allocated work items. The caller needs to wait until
185 * the wb threads have acked the work before it's safe to continue. 185 * the wb threads have acked the work before it's safe to continue.
186 */ 186 */
187 static void bdi_wait_on_work_clear(struct bdi_work *work) 187 static void bdi_wait_on_work_clear(struct bdi_work *work)
188 { 188 {
189 wait_on_bit(&work->state, WS_USED_B, bdi_sched_wait, 189 wait_on_bit(&work->state, WS_USED_B, bdi_sched_wait,
190 TASK_UNINTERRUPTIBLE); 190 TASK_UNINTERRUPTIBLE);
191 } 191 }
192 192
193 static void bdi_alloc_queue_work(struct backing_dev_info *bdi, 193 static void bdi_alloc_queue_work(struct backing_dev_info *bdi,
194 struct wb_writeback_args *args) 194 struct wb_writeback_args *args)
195 { 195 {
196 struct bdi_work *work; 196 struct bdi_work *work;
197 197
198 /* 198 /*
199 * This is WB_SYNC_NONE writeback, so if allocation fails just 199 * This is WB_SYNC_NONE writeback, so if allocation fails just
200 * wakeup the thread for old dirty data writeback 200 * wakeup the thread for old dirty data writeback
201 */ 201 */
202 work = kmalloc(sizeof(*work), GFP_ATOMIC); 202 work = kmalloc(sizeof(*work), GFP_ATOMIC);
203 if (work) { 203 if (work) {
204 bdi_work_init(work, args); 204 bdi_work_init(work, args);
205 bdi_queue_work(bdi, work); 205 bdi_queue_work(bdi, work);
206 } else { 206 } else {
207 struct bdi_writeback *wb = &bdi->wb; 207 struct bdi_writeback *wb = &bdi->wb;
208 208
209 if (wb->task) 209 if (wb->task)
210 wake_up_process(wb->task); 210 wake_up_process(wb->task);
211 } 211 }
212 } 212 }
213 213
214 /** 214 /**
215 * bdi_sync_writeback - start and wait for writeback 215 * bdi_sync_writeback - start and wait for writeback
216 * @bdi: the backing device to write from 216 * @bdi: the backing device to write from
217 * @sb: write inodes from this super_block 217 * @sb: write inodes from this super_block
218 * 218 *
219 * Description: 219 * Description:
220 * This does WB_SYNC_ALL data integrity writeback and waits for the 220 * This does WB_SYNC_ALL data integrity writeback and waits for the
221 * IO to complete. Callers must hold the sb s_umount semaphore for 221 * IO to complete. Callers must hold the sb s_umount semaphore for
222 * reading, to avoid having the super disappear before we are done. 222 * reading, to avoid having the super disappear before we are done.
223 */ 223 */
224 static void bdi_sync_writeback(struct backing_dev_info *bdi, 224 static void bdi_sync_writeback(struct backing_dev_info *bdi,
225 struct super_block *sb) 225 struct super_block *sb)
226 { 226 {
227 struct wb_writeback_args args = { 227 struct wb_writeback_args args = {
228 .sb = sb, 228 .sb = sb,
229 .sync_mode = WB_SYNC_ALL, 229 .sync_mode = WB_SYNC_ALL,
230 .nr_pages = LONG_MAX, 230 .nr_pages = LONG_MAX,
231 .range_cyclic = 0, 231 .range_cyclic = 0,
232 }; 232 };
233 struct bdi_work work; 233 struct bdi_work work;
234 234
235 bdi_work_init(&work, &args); 235 bdi_work_init(&work, &args);
236 work.state |= WS_ONSTACK; 236 work.state |= WS_ONSTACK;
237 237
238 bdi_queue_work(bdi, &work); 238 bdi_queue_work(bdi, &work);
239 bdi_wait_on_work_clear(&work); 239 bdi_wait_on_work_clear(&work);
240 } 240 }
241 241
242 /** 242 /**
243 * bdi_start_writeback - start writeback 243 * bdi_start_writeback - start writeback
244 * @bdi: the backing device to write from 244 * @bdi: the backing device to write from
245 * @nr_pages: the number of pages to write 245 * @nr_pages: the number of pages to write
246 * 246 *
247 * Description: 247 * Description:
248 * This does WB_SYNC_NONE opportunistic writeback. The IO is only 248 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
249 * started when this function returns, we make no guarentees on 249 * started when this function returns, we make no guarentees on
250 * completion. Caller need not hold sb s_umount semaphore. 250 * completion. Caller need not hold sb s_umount semaphore.
251 * 251 *
252 */ 252 */
253 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) 253 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages)
254 { 254 {
255 struct wb_writeback_args args = { 255 struct wb_writeback_args args = {
256 .sync_mode = WB_SYNC_NONE, 256 .sync_mode = WB_SYNC_NONE,
257 .nr_pages = nr_pages, 257 .nr_pages = nr_pages,
258 .range_cyclic = 1, 258 .range_cyclic = 1,
259 }; 259 };
260 260
261 /* 261 /*
262 * We treat @nr_pages=0 as the special case to do background writeback, 262 * We treat @nr_pages=0 as the special case to do background writeback,
263 * ie. to sync pages until the background dirty threshold is reached. 263 * ie. to sync pages until the background dirty threshold is reached.
264 */ 264 */
265 if (!nr_pages) { 265 if (!nr_pages) {
266 args.nr_pages = LONG_MAX; 266 args.nr_pages = LONG_MAX;
267 args.for_background = 1; 267 args.for_background = 1;
268 } 268 }
269 269
270 bdi_alloc_queue_work(bdi, &args); 270 bdi_alloc_queue_work(bdi, &args);
271 } 271 }
272 272
273 /* 273 /*
274 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 274 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
275 * furthest end of its superblock's dirty-inode list. 275 * furthest end of its superblock's dirty-inode list.
276 * 276 *
277 * Before stamping the inode's ->dirtied_when, we check to see whether it is 277 * Before stamping the inode's ->dirtied_when, we check to see whether it is
278 * already the most-recently-dirtied inode on the b_dirty list. If that is 278 * already the most-recently-dirtied inode on the b_dirty list. If that is
279 * the case then the inode must have been redirtied while it was being written 279 * the case then the inode must have been redirtied while it was being written
280 * out and we don't reset its dirtied_when. 280 * out and we don't reset its dirtied_when.
281 */ 281 */
282 static void redirty_tail(struct inode *inode) 282 static void redirty_tail(struct inode *inode)
283 { 283 {
284 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 284 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
285 285
286 if (!list_empty(&wb->b_dirty)) { 286 if (!list_empty(&wb->b_dirty)) {
287 struct inode *tail; 287 struct inode *tail;
288 288
289 tail = list_entry(wb->b_dirty.next, struct inode, i_list); 289 tail = list_entry(wb->b_dirty.next, struct inode, i_list);
290 if (time_before(inode->dirtied_when, tail->dirtied_when)) 290 if (time_before(inode->dirtied_when, tail->dirtied_when))
291 inode->dirtied_when = jiffies; 291 inode->dirtied_when = jiffies;
292 } 292 }
293 list_move(&inode->i_list, &wb->b_dirty); 293 list_move(&inode->i_list, &wb->b_dirty);
294 } 294 }
295 295
296 /* 296 /*
297 * requeue inode for re-scanning after bdi->b_io list is exhausted. 297 * requeue inode for re-scanning after bdi->b_io list is exhausted.
298 */ 298 */
299 static void requeue_io(struct inode *inode) 299 static void requeue_io(struct inode *inode)
300 { 300 {
301 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 301 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
302 302
303 list_move(&inode->i_list, &wb->b_more_io); 303 list_move(&inode->i_list, &wb->b_more_io);
304 } 304 }
305 305
306 static void inode_sync_complete(struct inode *inode) 306 static void inode_sync_complete(struct inode *inode)
307 { 307 {
308 /* 308 /*
309 * Prevent speculative execution through spin_unlock(&inode_lock); 309 * Prevent speculative execution through spin_unlock(&inode_lock);
310 */ 310 */
311 smp_mb(); 311 smp_mb();
312 wake_up_bit(&inode->i_state, __I_SYNC); 312 wake_up_bit(&inode->i_state, __I_SYNC);
313 } 313 }
314 314
315 static bool inode_dirtied_after(struct inode *inode, unsigned long t) 315 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
316 { 316 {
317 bool ret = time_after(inode->dirtied_when, t); 317 bool ret = time_after(inode->dirtied_when, t);
318 #ifndef CONFIG_64BIT 318 #ifndef CONFIG_64BIT
319 /* 319 /*
320 * For inodes being constantly redirtied, dirtied_when can get stuck. 320 * For inodes being constantly redirtied, dirtied_when can get stuck.
321 * It _appears_ to be in the future, but is actually in distant past. 321 * It _appears_ to be in the future, but is actually in distant past.
322 * This test is necessary to prevent such wrapped-around relative times 322 * This test is necessary to prevent such wrapped-around relative times
323 * from permanently stopping the whole pdflush writeback. 323 * from permanently stopping the whole pdflush writeback.
324 */ 324 */
325 ret = ret && time_before_eq(inode->dirtied_when, jiffies); 325 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
326 #endif 326 #endif
327 return ret; 327 return ret;
328 } 328 }
329 329
330 /* 330 /*
331 * Move expired dirty inodes from @delaying_queue to @dispatch_queue. 331 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
332 */ 332 */
333 static void move_expired_inodes(struct list_head *delaying_queue, 333 static void move_expired_inodes(struct list_head *delaying_queue,
334 struct list_head *dispatch_queue, 334 struct list_head *dispatch_queue,
335 unsigned long *older_than_this) 335 unsigned long *older_than_this)
336 { 336 {
337 while (!list_empty(delaying_queue)) { 337 while (!list_empty(delaying_queue)) {
338 struct inode *inode = list_entry(delaying_queue->prev, 338 struct inode *inode = list_entry(delaying_queue->prev,
339 struct inode, i_list); 339 struct inode, i_list);
340 if (older_than_this && 340 if (older_than_this &&
341 inode_dirtied_after(inode, *older_than_this)) 341 inode_dirtied_after(inode, *older_than_this))
342 break; 342 break;
343 list_move(&inode->i_list, dispatch_queue); 343 list_move(&inode->i_list, dispatch_queue);
344 } 344 }
345 } 345 }
346 346
347 /* 347 /*
348 * Queue all expired dirty inodes for io, eldest first. 348 * Queue all expired dirty inodes for io, eldest first.
349 */ 349 */
350 static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) 350 static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
351 { 351 {
352 list_splice_init(&wb->b_more_io, wb->b_io.prev); 352 list_splice_init(&wb->b_more_io, wb->b_io.prev);
353 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); 353 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
354 } 354 }
355 355
356 static int write_inode(struct inode *inode, int sync) 356 static int write_inode(struct inode *inode, int sync)
357 { 357 {
358 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) 358 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
359 return inode->i_sb->s_op->write_inode(inode, sync); 359 return inode->i_sb->s_op->write_inode(inode, sync);
360 return 0; 360 return 0;
361 } 361 }
362 362
363 /* 363 /*
364 * Wait for writeback on an inode to complete. 364 * Wait for writeback on an inode to complete.
365 */ 365 */
366 static void inode_wait_for_writeback(struct inode *inode) 366 static void inode_wait_for_writeback(struct inode *inode)
367 { 367 {
368 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 368 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
369 wait_queue_head_t *wqh; 369 wait_queue_head_t *wqh;
370 370
371 wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 371 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
372 do { 372 do {
373 spin_unlock(&inode_lock); 373 spin_unlock(&inode_lock);
374 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 374 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
375 spin_lock(&inode_lock); 375 spin_lock(&inode_lock);
376 } while (inode->i_state & I_SYNC); 376 } while (inode->i_state & I_SYNC);
377 } 377 }
378 378
379 /* 379 /*
380 * Write out an inode's dirty pages. Called under inode_lock. Either the 380 * Write out an inode's dirty pages. Called under inode_lock. Either the
381 * caller has ref on the inode (either via __iget or via syscall against an fd) 381 * caller has ref on the inode (either via __iget or via syscall against an fd)
382 * or the inode has I_WILL_FREE set (via generic_forget_inode) 382 * or the inode has I_WILL_FREE set (via generic_forget_inode)
383 * 383 *
384 * If `wait' is set, wait on the writeout. 384 * If `wait' is set, wait on the writeout.
385 * 385 *
386 * The whole writeout design is quite complex and fragile. We want to avoid 386 * The whole writeout design is quite complex and fragile. We want to avoid
387 * starvation of particular inodes when others are being redirtied, prevent 387 * starvation of particular inodes when others are being redirtied, prevent
388 * livelocks, etc. 388 * livelocks, etc.
389 * 389 *
390 * Called under inode_lock. 390 * Called under inode_lock.
391 */ 391 */
392 static int 392 static int
393 writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 393 writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
394 { 394 {
395 struct address_space *mapping = inode->i_mapping; 395 struct address_space *mapping = inode->i_mapping;
396 int wait = wbc->sync_mode == WB_SYNC_ALL; 396 int wait = wbc->sync_mode == WB_SYNC_ALL;
397 unsigned dirty; 397 unsigned dirty;
398 int ret; 398 int ret;
399 399
400 if (!atomic_read(&inode->i_count)) 400 if (!atomic_read(&inode->i_count))
401 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 401 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
402 else 402 else
403 WARN_ON(inode->i_state & I_WILL_FREE); 403 WARN_ON(inode->i_state & I_WILL_FREE);
404 404
405 if (inode->i_state & I_SYNC) { 405 if (inode->i_state & I_SYNC) {
406 /* 406 /*
407 * If this inode is locked for writeback and we are not doing 407 * If this inode is locked for writeback and we are not doing
408 * writeback-for-data-integrity, move it to b_more_io so that 408 * writeback-for-data-integrity, move it to b_more_io so that
409 * writeback can proceed with the other inodes on s_io. 409 * writeback can proceed with the other inodes on s_io.
410 * 410 *
411 * We'll have another go at writing back this inode when we 411 * We'll have another go at writing back this inode when we
412 * completed a full scan of b_io. 412 * completed a full scan of b_io.
413 */ 413 */
414 if (!wait) { 414 if (!wait) {
415 requeue_io(inode); 415 requeue_io(inode);
416 return 0; 416 return 0;
417 } 417 }
418 418
419 /* 419 /*
420 * It's a data-integrity sync. We must wait. 420 * It's a data-integrity sync. We must wait.
421 */ 421 */
422 inode_wait_for_writeback(inode); 422 inode_wait_for_writeback(inode);
423 } 423 }
424 424
425 BUG_ON(inode->i_state & I_SYNC); 425 BUG_ON(inode->i_state & I_SYNC);
426 426
427 /* Set I_SYNC, reset I_DIRTY */ 427 /* Set I_SYNC, reset I_DIRTY */
428 dirty = inode->i_state & I_DIRTY; 428 dirty = inode->i_state & I_DIRTY;
429 inode->i_state |= I_SYNC; 429 inode->i_state |= I_SYNC;
430 inode->i_state &= ~I_DIRTY; 430 inode->i_state &= ~I_DIRTY;
431 431
432 spin_unlock(&inode_lock); 432 spin_unlock(&inode_lock);
433 433
434 ret = do_writepages(mapping, wbc); 434 ret = do_writepages(mapping, wbc);
435 435
436 /* Don't write the inode if only I_DIRTY_PAGES was set */ 436 /* Don't write the inode if only I_DIRTY_PAGES was set */
437 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 437 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
438 int err = write_inode(inode, wait); 438 int err = write_inode(inode, wait);
439 if (ret == 0) 439 if (ret == 0)
440 ret = err; 440 ret = err;
441 } 441 }
442 442
443 if (wait) { 443 if (wait) {
444 int err = filemap_fdatawait(mapping); 444 int err = filemap_fdatawait(mapping);
445 if (ret == 0) 445 if (ret == 0)
446 ret = err; 446 ret = err;
447 } 447 }
448 448
449 spin_lock(&inode_lock); 449 spin_lock(&inode_lock);
450 inode->i_state &= ~I_SYNC; 450 inode->i_state &= ~I_SYNC;
451 if (!(inode->i_state & (I_FREEING | I_CLEAR))) { 451 if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
452 if (!(inode->i_state & I_DIRTY) && 452 if (inode->i_state & I_DIRTY) {
453 mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
454 /* 453 /*
454 * Someone redirtied the inode while were writing back
455 * the pages.
456 */
457 redirty_tail(inode);
458 } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
459 /*
455 * We didn't write back all the pages. nfs_writepages() 460 * We didn't write back all the pages. nfs_writepages()
456 * sometimes bales out without doing anything. Redirty 461 * sometimes bales out without doing anything. Redirty
457 * the inode; Move it from b_io onto b_more_io/b_dirty. 462 * the inode; Move it from b_io onto b_more_io/b_dirty.
458 */ 463 */
459 /* 464 /*
460 * akpm: if the caller was the kupdate function we put 465 * akpm: if the caller was the kupdate function we put
461 * this inode at the head of b_dirty so it gets first 466 * this inode at the head of b_dirty so it gets first
462 * consideration. Otherwise, move it to the tail, for 467 * consideration. Otherwise, move it to the tail, for
463 * the reasons described there. I'm not really sure 468 * the reasons described there. I'm not really sure
464 * how much sense this makes. Presumably I had a good 469 * how much sense this makes. Presumably I had a good
465 * reasons for doing it this way, and I'd rather not 470 * reasons for doing it this way, and I'd rather not
466 * muck with it at present. 471 * muck with it at present.
467 */ 472 */
468 if (wbc->for_kupdate) { 473 if (wbc->for_kupdate) {
469 /* 474 /*
470 * For the kupdate function we move the inode 475 * For the kupdate function we move the inode
471 * to b_more_io so it will get more writeout as 476 * to b_more_io so it will get more writeout as
472 * soon as the queue becomes uncongested. 477 * soon as the queue becomes uncongested.
473 */ 478 */
474 inode->i_state |= I_DIRTY_PAGES; 479 inode->i_state |= I_DIRTY_PAGES;
475 if (wbc->nr_to_write <= 0) { 480 if (wbc->nr_to_write <= 0) {
476 /* 481 /*
477 * slice used up: queue for next turn 482 * slice used up: queue for next turn
478 */ 483 */
479 requeue_io(inode); 484 requeue_io(inode);
480 } else { 485 } else {
481 /* 486 /*
482 * somehow blocked: retry later 487 * somehow blocked: retry later
483 */ 488 */
484 redirty_tail(inode); 489 redirty_tail(inode);
485 } 490 }
486 } else { 491 } else {
487 /* 492 /*
488 * Otherwise fully redirty the inode so that 493 * Otherwise fully redirty the inode so that
489 * other inodes on this superblock will get some 494 * other inodes on this superblock will get some
490 * writeout. Otherwise heavy writing to one 495 * writeout. Otherwise heavy writing to one
491 * file would indefinitely suspend writeout of 496 * file would indefinitely suspend writeout of
492 * all the other files. 497 * all the other files.
493 */ 498 */
494 inode->i_state |= I_DIRTY_PAGES; 499 inode->i_state |= I_DIRTY_PAGES;
495 redirty_tail(inode); 500 redirty_tail(inode);
496 } 501 }
497 } else if (inode->i_state & I_DIRTY) {
498 /*
499 * Someone redirtied the inode while were writing back
500 * the pages.
501 */
502 redirty_tail(inode);
503 } else if (atomic_read(&inode->i_count)) { 502 } else if (atomic_read(&inode->i_count)) {
504 /* 503 /*
505 * The inode is clean, inuse 504 * The inode is clean, inuse
506 */ 505 */
507 list_move(&inode->i_list, &inode_in_use); 506 list_move(&inode->i_list, &inode_in_use);
508 } else { 507 } else {
509 /* 508 /*
510 * The inode is clean, unused 509 * The inode is clean, unused
511 */ 510 */
512 list_move(&inode->i_list, &inode_unused); 511 list_move(&inode->i_list, &inode_unused);
513 } 512 }
514 } 513 }
515 inode_sync_complete(inode); 514 inode_sync_complete(inode);
516 return ret; 515 return ret;
517 } 516 }
518 517
519 /* 518 /*
520 * For WB_SYNC_NONE writeback, the caller does not have the sb pinned 519 * For WB_SYNC_NONE writeback, the caller does not have the sb pinned
521 * before calling writeback. So make sure that we do pin it, so it doesn't 520 * before calling writeback. So make sure that we do pin it, so it doesn't
522 * go away while we are writing inodes from it. 521 * go away while we are writing inodes from it.
523 * 522 *
524 * Returns 0 if the super was successfully pinned (or pinning wasn't needed), 523 * Returns 0 if the super was successfully pinned (or pinning wasn't needed),
525 * 1 if we failed. 524 * 1 if we failed.
526 */ 525 */
527 static int pin_sb_for_writeback(struct writeback_control *wbc, 526 static int pin_sb_for_writeback(struct writeback_control *wbc,
528 struct inode *inode) 527 struct inode *inode)
529 { 528 {
530 struct super_block *sb = inode->i_sb; 529 struct super_block *sb = inode->i_sb;
531 530
532 /* 531 /*
533 * Caller must already hold the ref for this 532 * Caller must already hold the ref for this
534 */ 533 */
535 if (wbc->sync_mode == WB_SYNC_ALL) { 534 if (wbc->sync_mode == WB_SYNC_ALL) {
536 WARN_ON(!rwsem_is_locked(&sb->s_umount)); 535 WARN_ON(!rwsem_is_locked(&sb->s_umount));
537 return 0; 536 return 0;
538 } 537 }
539 538
540 spin_lock(&sb_lock); 539 spin_lock(&sb_lock);
541 sb->s_count++; 540 sb->s_count++;
542 if (down_read_trylock(&sb->s_umount)) { 541 if (down_read_trylock(&sb->s_umount)) {
543 if (sb->s_root) { 542 if (sb->s_root) {
544 spin_unlock(&sb_lock); 543 spin_unlock(&sb_lock);
545 return 0; 544 return 0;
546 } 545 }
547 /* 546 /*
548 * umounted, drop rwsem again and fall through to failure 547 * umounted, drop rwsem again and fall through to failure
549 */ 548 */
550 up_read(&sb->s_umount); 549 up_read(&sb->s_umount);
551 } 550 }
552 551
553 sb->s_count--; 552 sb->s_count--;
554 spin_unlock(&sb_lock); 553 spin_unlock(&sb_lock);
555 return 1; 554 return 1;
556 } 555 }
557 556
558 static void unpin_sb_for_writeback(struct writeback_control *wbc, 557 static void unpin_sb_for_writeback(struct writeback_control *wbc,
559 struct inode *inode) 558 struct inode *inode)
560 { 559 {
561 struct super_block *sb = inode->i_sb; 560 struct super_block *sb = inode->i_sb;
562 561
563 if (wbc->sync_mode == WB_SYNC_ALL) 562 if (wbc->sync_mode == WB_SYNC_ALL)
564 return; 563 return;
565 564
566 up_read(&sb->s_umount); 565 up_read(&sb->s_umount);
567 put_super(sb); 566 put_super(sb);
568 } 567 }
569 568
570 static void writeback_inodes_wb(struct bdi_writeback *wb, 569 static void writeback_inodes_wb(struct bdi_writeback *wb,
571 struct writeback_control *wbc) 570 struct writeback_control *wbc)
572 { 571 {
573 struct super_block *sb = wbc->sb; 572 struct super_block *sb = wbc->sb;
574 const int is_blkdev_sb = sb_is_blkdev_sb(sb); 573 const int is_blkdev_sb = sb_is_blkdev_sb(sb);
575 const unsigned long start = jiffies; /* livelock avoidance */ 574 const unsigned long start = jiffies; /* livelock avoidance */
576 575
577 spin_lock(&inode_lock); 576 spin_lock(&inode_lock);
578 577
579 if (!wbc->for_kupdate || list_empty(&wb->b_io)) 578 if (!wbc->for_kupdate || list_empty(&wb->b_io))
580 queue_io(wb, wbc->older_than_this); 579 queue_io(wb, wbc->older_than_this);
581 580
582 while (!list_empty(&wb->b_io)) { 581 while (!list_empty(&wb->b_io)) {
583 struct inode *inode = list_entry(wb->b_io.prev, 582 struct inode *inode = list_entry(wb->b_io.prev,
584 struct inode, i_list); 583 struct inode, i_list);
585 long pages_skipped; 584 long pages_skipped;
586 585
587 /* 586 /*
588 * super block given and doesn't match, skip this inode 587 * super block given and doesn't match, skip this inode
589 */ 588 */
590 if (sb && sb != inode->i_sb) { 589 if (sb && sb != inode->i_sb) {
591 redirty_tail(inode); 590 redirty_tail(inode);
592 continue; 591 continue;
593 } 592 }
594 593
595 if (!bdi_cap_writeback_dirty(wb->bdi)) { 594 if (!bdi_cap_writeback_dirty(wb->bdi)) {
596 redirty_tail(inode); 595 redirty_tail(inode);
597 if (is_blkdev_sb) { 596 if (is_blkdev_sb) {
598 /* 597 /*
599 * Dirty memory-backed blockdev: the ramdisk 598 * Dirty memory-backed blockdev: the ramdisk
600 * driver does this. Skip just this inode 599 * driver does this. Skip just this inode
601 */ 600 */
602 continue; 601 continue;
603 } 602 }
604 /* 603 /*
605 * Dirty memory-backed inode against a filesystem other 604 * Dirty memory-backed inode against a filesystem other
606 * than the kernel-internal bdev filesystem. Skip the 605 * than the kernel-internal bdev filesystem. Skip the
607 * entire superblock. 606 * entire superblock.
608 */ 607 */
609 break; 608 break;
610 } 609 }
611 610
612 if (inode->i_state & (I_NEW | I_WILL_FREE)) { 611 if (inode->i_state & (I_NEW | I_WILL_FREE)) {
613 requeue_io(inode); 612 requeue_io(inode);
614 continue; 613 continue;
615 } 614 }
616 615
617 if (wbc->nonblocking && bdi_write_congested(wb->bdi)) { 616 if (wbc->nonblocking && bdi_write_congested(wb->bdi)) {
618 wbc->encountered_congestion = 1; 617 wbc->encountered_congestion = 1;
619 if (!is_blkdev_sb) 618 if (!is_blkdev_sb)
620 break; /* Skip a congested fs */ 619 break; /* Skip a congested fs */
621 requeue_io(inode); 620 requeue_io(inode);
622 continue; /* Skip a congested blockdev */ 621 continue; /* Skip a congested blockdev */
623 } 622 }
624 623
625 /* 624 /*
626 * Was this inode dirtied after sync_sb_inodes was called? 625 * Was this inode dirtied after sync_sb_inodes was called?
627 * This keeps sync from extra jobs and livelock. 626 * This keeps sync from extra jobs and livelock.
628 */ 627 */
629 if (inode_dirtied_after(inode, start)) 628 if (inode_dirtied_after(inode, start))
630 break; 629 break;
631 630
632 if (pin_sb_for_writeback(wbc, inode)) { 631 if (pin_sb_for_writeback(wbc, inode)) {
633 requeue_io(inode); 632 requeue_io(inode);
634 continue; 633 continue;
635 } 634 }
636 635
637 BUG_ON(inode->i_state & (I_FREEING | I_CLEAR)); 636 BUG_ON(inode->i_state & (I_FREEING | I_CLEAR));
638 __iget(inode); 637 __iget(inode);
639 pages_skipped = wbc->pages_skipped; 638 pages_skipped = wbc->pages_skipped;
640 writeback_single_inode(inode, wbc); 639 writeback_single_inode(inode, wbc);
641 unpin_sb_for_writeback(wbc, inode); 640 unpin_sb_for_writeback(wbc, inode);
642 if (wbc->pages_skipped != pages_skipped) { 641 if (wbc->pages_skipped != pages_skipped) {
643 /* 642 /*
644 * writeback is not making progress due to locked 643 * writeback is not making progress due to locked
645 * buffers. Skip this inode for now. 644 * buffers. Skip this inode for now.
646 */ 645 */
647 redirty_tail(inode); 646 redirty_tail(inode);
648 } 647 }
649 spin_unlock(&inode_lock); 648 spin_unlock(&inode_lock);
650 iput(inode); 649 iput(inode);
651 cond_resched(); 650 cond_resched();
652 spin_lock(&inode_lock); 651 spin_lock(&inode_lock);
653 if (wbc->nr_to_write <= 0) { 652 if (wbc->nr_to_write <= 0) {
654 wbc->more_io = 1; 653 wbc->more_io = 1;
655 break; 654 break;
656 } 655 }
657 if (!list_empty(&wb->b_more_io)) 656 if (!list_empty(&wb->b_more_io))
658 wbc->more_io = 1; 657 wbc->more_io = 1;
659 } 658 }
660 659
661 spin_unlock(&inode_lock); 660 spin_unlock(&inode_lock);
662 /* Leave any unwritten inodes on b_io */ 661 /* Leave any unwritten inodes on b_io */
663 } 662 }
664 663
665 void writeback_inodes_wbc(struct writeback_control *wbc) 664 void writeback_inodes_wbc(struct writeback_control *wbc)
666 { 665 {
667 struct backing_dev_info *bdi = wbc->bdi; 666 struct backing_dev_info *bdi = wbc->bdi;
668 667
669 writeback_inodes_wb(&bdi->wb, wbc); 668 writeback_inodes_wb(&bdi->wb, wbc);
670 } 669 }
671 670
672 /* 671 /*
673 * The maximum number of pages to writeout in a single bdi flush/kupdate 672 * The maximum number of pages to writeout in a single bdi flush/kupdate
674 * operation. We do this so we don't hold I_SYNC against an inode for 673 * operation. We do this so we don't hold I_SYNC against an inode for
675 * enormous amounts of time, which would block a userspace task which has 674 * enormous amounts of time, which would block a userspace task which has
676 * been forced to throttle against that inode. Also, the code reevaluates 675 * been forced to throttle against that inode. Also, the code reevaluates
677 * the dirty each time it has written this many pages. 676 * the dirty each time it has written this many pages.
678 */ 677 */
679 #define MAX_WRITEBACK_PAGES 1024 678 #define MAX_WRITEBACK_PAGES 1024
680 679
681 static inline bool over_bground_thresh(void) 680 static inline bool over_bground_thresh(void)
682 { 681 {
683 unsigned long background_thresh, dirty_thresh; 682 unsigned long background_thresh, dirty_thresh;
684 683
685 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); 684 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL);
686 685
687 return (global_page_state(NR_FILE_DIRTY) + 686 return (global_page_state(NR_FILE_DIRTY) +
688 global_page_state(NR_UNSTABLE_NFS) >= background_thresh); 687 global_page_state(NR_UNSTABLE_NFS) >= background_thresh);
689 } 688 }
690 689
691 /* 690 /*
692 * Explicit flushing or periodic writeback of "old" data. 691 * Explicit flushing or periodic writeback of "old" data.
693 * 692 *
694 * Define "old": the first time one of an inode's pages is dirtied, we mark the 693 * Define "old": the first time one of an inode's pages is dirtied, we mark the
695 * dirtying-time in the inode's address_space. So this periodic writeback code 694 * dirtying-time in the inode's address_space. So this periodic writeback code
696 * just walks the superblock inode list, writing back any inodes which are 695 * just walks the superblock inode list, writing back any inodes which are
697 * older than a specific point in time. 696 * older than a specific point in time.
698 * 697 *
699 * Try to run once per dirty_writeback_interval. But if a writeback event 698 * Try to run once per dirty_writeback_interval. But if a writeback event
700 * takes longer than a dirty_writeback_interval interval, then leave a 699 * takes longer than a dirty_writeback_interval interval, then leave a
701 * one-second gap. 700 * one-second gap.
702 * 701 *
703 * older_than_this takes precedence over nr_to_write. So we'll only write back 702 * older_than_this takes precedence over nr_to_write. So we'll only write back
704 * all dirty pages if they are all attached to "old" mappings. 703 * all dirty pages if they are all attached to "old" mappings.
705 */ 704 */
706 static long wb_writeback(struct bdi_writeback *wb, 705 static long wb_writeback(struct bdi_writeback *wb,
707 struct wb_writeback_args *args) 706 struct wb_writeback_args *args)
708 { 707 {
709 struct writeback_control wbc = { 708 struct writeback_control wbc = {
710 .bdi = wb->bdi, 709 .bdi = wb->bdi,
711 .sb = args->sb, 710 .sb = args->sb,
712 .sync_mode = args->sync_mode, 711 .sync_mode = args->sync_mode,
713 .older_than_this = NULL, 712 .older_than_this = NULL,
714 .for_kupdate = args->for_kupdate, 713 .for_kupdate = args->for_kupdate,
715 .range_cyclic = args->range_cyclic, 714 .range_cyclic = args->range_cyclic,
716 }; 715 };
717 unsigned long oldest_jif; 716 unsigned long oldest_jif;
718 long wrote = 0; 717 long wrote = 0;
719 struct inode *inode; 718 struct inode *inode;
720 719
721 if (wbc.for_kupdate) { 720 if (wbc.for_kupdate) {
722 wbc.older_than_this = &oldest_jif; 721 wbc.older_than_this = &oldest_jif;
723 oldest_jif = jiffies - 722 oldest_jif = jiffies -
724 msecs_to_jiffies(dirty_expire_interval * 10); 723 msecs_to_jiffies(dirty_expire_interval * 10);
725 } 724 }
726 if (!wbc.range_cyclic) { 725 if (!wbc.range_cyclic) {
727 wbc.range_start = 0; 726 wbc.range_start = 0;
728 wbc.range_end = LLONG_MAX; 727 wbc.range_end = LLONG_MAX;
729 } 728 }
730 729
731 for (;;) { 730 for (;;) {
732 /* 731 /*
733 * Stop writeback when nr_pages has been consumed 732 * Stop writeback when nr_pages has been consumed
734 */ 733 */
735 if (args->nr_pages <= 0) 734 if (args->nr_pages <= 0)
736 break; 735 break;
737 736
738 /* 737 /*
739 * For background writeout, stop when we are below the 738 * For background writeout, stop when we are below the
740 * background dirty threshold 739 * background dirty threshold
741 */ 740 */
742 if (args->for_background && !over_bground_thresh()) 741 if (args->for_background && !over_bground_thresh())
743 break; 742 break;
744 743
745 wbc.more_io = 0; 744 wbc.more_io = 0;
746 wbc.encountered_congestion = 0; 745 wbc.encountered_congestion = 0;
747 wbc.nr_to_write = MAX_WRITEBACK_PAGES; 746 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
748 wbc.pages_skipped = 0; 747 wbc.pages_skipped = 0;
749 writeback_inodes_wb(wb, &wbc); 748 writeback_inodes_wb(wb, &wbc);
750 args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; 749 args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
751 wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; 750 wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write;
752 751
753 /* 752 /*
754 * If we ran out of stuff to write, bail unless more_io got set 753 * If we ran out of stuff to write, bail unless more_io got set
755 */ 754 */
756 if (wbc.nr_to_write > 0) { 755 if (wbc.nr_to_write > 0) {
757 if (wbc.more_io) { 756 if (wbc.more_io) {
758 if (wbc.nr_to_write < MAX_WRITEBACK_PAGES) 757 if (wbc.nr_to_write < MAX_WRITEBACK_PAGES)
759 continue; 758 continue;
760 /* 759 /*
761 * Nothing written. Wait for some inode to 760 * Nothing written. Wait for some inode to
762 * become available for writeback. Otherwise 761 * become available for writeback. Otherwise
763 * we'll just busyloop. 762 * we'll just busyloop.
764 */ 763 */
765 spin_lock(&inode_lock); 764 spin_lock(&inode_lock);
766 if (!list_empty(&wb->b_more_io)) { 765 if (!list_empty(&wb->b_more_io)) {
767 inode = list_entry( 766 inode = list_entry(
768 wb->b_more_io.prev, 767 wb->b_more_io.prev,
769 struct inode, i_list); 768 struct inode, i_list);
770 inode_wait_for_writeback(inode); 769 inode_wait_for_writeback(inode);
771 } 770 }
772 spin_unlock(&inode_lock); 771 spin_unlock(&inode_lock);
773 continue; 772 continue;
774 } 773 }
775 break; 774 break;
776 } 775 }
777 } 776 }
778 777
779 return wrote; 778 return wrote;
780 } 779 }
781 780
782 /* 781 /*
783 * Return the next bdi_work struct that hasn't been processed by this 782 * Return the next bdi_work struct that hasn't been processed by this
784 * wb thread yet. ->seen is initially set for each thread that exists 783 * wb thread yet. ->seen is initially set for each thread that exists
785 * for this device, when a thread first notices a piece of work it 784 * for this device, when a thread first notices a piece of work it
786 * clears its bit. Depending on writeback type, the thread will notify 785 * clears its bit. Depending on writeback type, the thread will notify
787 * completion on either receiving the work (WB_SYNC_NONE) or after 786 * completion on either receiving the work (WB_SYNC_NONE) or after
788 * it is done (WB_SYNC_ALL). 787 * it is done (WB_SYNC_ALL).
789 */ 788 */
790 static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi, 789 static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi,
791 struct bdi_writeback *wb) 790 struct bdi_writeback *wb)
792 { 791 {
793 struct bdi_work *work, *ret = NULL; 792 struct bdi_work *work, *ret = NULL;
794 793
795 rcu_read_lock(); 794 rcu_read_lock();
796 795
797 list_for_each_entry_rcu(work, &bdi->work_list, list) { 796 list_for_each_entry_rcu(work, &bdi->work_list, list) {
798 if (!test_bit(wb->nr, &work->seen)) 797 if (!test_bit(wb->nr, &work->seen))
799 continue; 798 continue;
800 clear_bit(wb->nr, &work->seen); 799 clear_bit(wb->nr, &work->seen);
801 800
802 ret = work; 801 ret = work;
803 break; 802 break;
804 } 803 }
805 804
806 rcu_read_unlock(); 805 rcu_read_unlock();
807 return ret; 806 return ret;
808 } 807 }
809 808
810 static long wb_check_old_data_flush(struct bdi_writeback *wb) 809 static long wb_check_old_data_flush(struct bdi_writeback *wb)
811 { 810 {
812 unsigned long expired; 811 unsigned long expired;
813 long nr_pages; 812 long nr_pages;
814 813
815 expired = wb->last_old_flush + 814 expired = wb->last_old_flush +
816 msecs_to_jiffies(dirty_writeback_interval * 10); 815 msecs_to_jiffies(dirty_writeback_interval * 10);
817 if (time_before(jiffies, expired)) 816 if (time_before(jiffies, expired))
818 return 0; 817 return 0;
819 818
820 wb->last_old_flush = jiffies; 819 wb->last_old_flush = jiffies;
821 nr_pages = global_page_state(NR_FILE_DIRTY) + 820 nr_pages = global_page_state(NR_FILE_DIRTY) +
822 global_page_state(NR_UNSTABLE_NFS) + 821 global_page_state(NR_UNSTABLE_NFS) +
823 (inodes_stat.nr_inodes - inodes_stat.nr_unused); 822 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
824 823
825 if (nr_pages) { 824 if (nr_pages) {
826 struct wb_writeback_args args = { 825 struct wb_writeback_args args = {
827 .nr_pages = nr_pages, 826 .nr_pages = nr_pages,
828 .sync_mode = WB_SYNC_NONE, 827 .sync_mode = WB_SYNC_NONE,
829 .for_kupdate = 1, 828 .for_kupdate = 1,
830 .range_cyclic = 1, 829 .range_cyclic = 1,
831 }; 830 };
832 831
833 return wb_writeback(wb, &args); 832 return wb_writeback(wb, &args);
834 } 833 }
835 834
836 return 0; 835 return 0;
837 } 836 }
838 837
839 /* 838 /*
840 * Retrieve work items and do the writeback they describe 839 * Retrieve work items and do the writeback they describe
841 */ 840 */
842 long wb_do_writeback(struct bdi_writeback *wb, int force_wait) 841 long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
843 { 842 {
844 struct backing_dev_info *bdi = wb->bdi; 843 struct backing_dev_info *bdi = wb->bdi;
845 struct bdi_work *work; 844 struct bdi_work *work;
846 long wrote = 0; 845 long wrote = 0;
847 846
848 while ((work = get_next_work_item(bdi, wb)) != NULL) { 847 while ((work = get_next_work_item(bdi, wb)) != NULL) {
849 struct wb_writeback_args args = work->args; 848 struct wb_writeback_args args = work->args;
850 849
851 /* 850 /*
852 * Override sync mode, in case we must wait for completion 851 * Override sync mode, in case we must wait for completion
853 */ 852 */
854 if (force_wait) 853 if (force_wait)
855 work->args.sync_mode = args.sync_mode = WB_SYNC_ALL; 854 work->args.sync_mode = args.sync_mode = WB_SYNC_ALL;
856 855
857 /* 856 /*
858 * If this isn't a data integrity operation, just notify 857 * If this isn't a data integrity operation, just notify
859 * that we have seen this work and we are now starting it. 858 * that we have seen this work and we are now starting it.
860 */ 859 */
861 if (args.sync_mode == WB_SYNC_NONE) 860 if (args.sync_mode == WB_SYNC_NONE)
862 wb_clear_pending(wb, work); 861 wb_clear_pending(wb, work);
863 862
864 wrote += wb_writeback(wb, &args); 863 wrote += wb_writeback(wb, &args);
865 864
866 /* 865 /*
867 * This is a data integrity writeback, so only do the 866 * This is a data integrity writeback, so only do the
868 * notification when we have completed the work. 867 * notification when we have completed the work.
869 */ 868 */
870 if (args.sync_mode == WB_SYNC_ALL) 869 if (args.sync_mode == WB_SYNC_ALL)
871 wb_clear_pending(wb, work); 870 wb_clear_pending(wb, work);
872 } 871 }
873 872
874 /* 873 /*
875 * Check for periodic writeback, kupdated() style 874 * Check for periodic writeback, kupdated() style
876 */ 875 */
877 wrote += wb_check_old_data_flush(wb); 876 wrote += wb_check_old_data_flush(wb);
878 877
879 return wrote; 878 return wrote;
880 } 879 }
881 880
882 /* 881 /*
883 * Handle writeback of dirty data for the device backed by this bdi. Also 882 * Handle writeback of dirty data for the device backed by this bdi. Also
884 * wakes up periodically and does kupdated style flushing. 883 * wakes up periodically and does kupdated style flushing.
885 */ 884 */
886 int bdi_writeback_task(struct bdi_writeback *wb) 885 int bdi_writeback_task(struct bdi_writeback *wb)
887 { 886 {
888 unsigned long last_active = jiffies; 887 unsigned long last_active = jiffies;
889 unsigned long wait_jiffies = -1UL; 888 unsigned long wait_jiffies = -1UL;
890 long pages_written; 889 long pages_written;
891 890
892 while (!kthread_should_stop()) { 891 while (!kthread_should_stop()) {
893 pages_written = wb_do_writeback(wb, 0); 892 pages_written = wb_do_writeback(wb, 0);
894 893
895 if (pages_written) 894 if (pages_written)
896 last_active = jiffies; 895 last_active = jiffies;
897 else if (wait_jiffies != -1UL) { 896 else if (wait_jiffies != -1UL) {
898 unsigned long max_idle; 897 unsigned long max_idle;
899 898
900 /* 899 /*
901 * Longest period of inactivity that we tolerate. If we 900 * Longest period of inactivity that we tolerate. If we
902 * see dirty data again later, the task will get 901 * see dirty data again later, the task will get
903 * recreated automatically. 902 * recreated automatically.
904 */ 903 */
905 max_idle = max(5UL * 60 * HZ, wait_jiffies); 904 max_idle = max(5UL * 60 * HZ, wait_jiffies);
906 if (time_after(jiffies, max_idle + last_active)) 905 if (time_after(jiffies, max_idle + last_active))
907 break; 906 break;
908 } 907 }
909 908
910 wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10); 909 wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10);
911 schedule_timeout_interruptible(wait_jiffies); 910 schedule_timeout_interruptible(wait_jiffies);
912 try_to_freeze(); 911 try_to_freeze();
913 } 912 }
914 913
915 return 0; 914 return 0;
916 } 915 }
917 916
918 /* 917 /*
919 * Schedule writeback for all backing devices. This does WB_SYNC_NONE 918 * Schedule writeback for all backing devices. This does WB_SYNC_NONE
920 * writeback, for integrity writeback see bdi_sync_writeback(). 919 * writeback, for integrity writeback see bdi_sync_writeback().
921 */ 920 */
922 static void bdi_writeback_all(struct super_block *sb, long nr_pages) 921 static void bdi_writeback_all(struct super_block *sb, long nr_pages)
923 { 922 {
924 struct wb_writeback_args args = { 923 struct wb_writeback_args args = {
925 .sb = sb, 924 .sb = sb,
926 .nr_pages = nr_pages, 925 .nr_pages = nr_pages,
927 .sync_mode = WB_SYNC_NONE, 926 .sync_mode = WB_SYNC_NONE,
928 }; 927 };
929 struct backing_dev_info *bdi; 928 struct backing_dev_info *bdi;
930 929
931 rcu_read_lock(); 930 rcu_read_lock();
932 931
933 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 932 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
934 if (!bdi_has_dirty_io(bdi)) 933 if (!bdi_has_dirty_io(bdi))
935 continue; 934 continue;
936 935
937 bdi_alloc_queue_work(bdi, &args); 936 bdi_alloc_queue_work(bdi, &args);
938 } 937 }
939 938
940 rcu_read_unlock(); 939 rcu_read_unlock();
941 } 940 }
942 941
943 /* 942 /*
944 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 943 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
945 * the whole world. 944 * the whole world.
946 */ 945 */
947 void wakeup_flusher_threads(long nr_pages) 946 void wakeup_flusher_threads(long nr_pages)
948 { 947 {
949 if (nr_pages == 0) 948 if (nr_pages == 0)
950 nr_pages = global_page_state(NR_FILE_DIRTY) + 949 nr_pages = global_page_state(NR_FILE_DIRTY) +
951 global_page_state(NR_UNSTABLE_NFS); 950 global_page_state(NR_UNSTABLE_NFS);
952 bdi_writeback_all(NULL, nr_pages); 951 bdi_writeback_all(NULL, nr_pages);
953 } 952 }
954 953
955 static noinline void block_dump___mark_inode_dirty(struct inode *inode) 954 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
956 { 955 {
957 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 956 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
958 struct dentry *dentry; 957 struct dentry *dentry;
959 const char *name = "?"; 958 const char *name = "?";
960 959
961 dentry = d_find_alias(inode); 960 dentry = d_find_alias(inode);
962 if (dentry) { 961 if (dentry) {
963 spin_lock(&dentry->d_lock); 962 spin_lock(&dentry->d_lock);
964 name = (const char *) dentry->d_name.name; 963 name = (const char *) dentry->d_name.name;
965 } 964 }
966 printk(KERN_DEBUG 965 printk(KERN_DEBUG
967 "%s(%d): dirtied inode %lu (%s) on %s\n", 966 "%s(%d): dirtied inode %lu (%s) on %s\n",
968 current->comm, task_pid_nr(current), inode->i_ino, 967 current->comm, task_pid_nr(current), inode->i_ino,
969 name, inode->i_sb->s_id); 968 name, inode->i_sb->s_id);
970 if (dentry) { 969 if (dentry) {
971 spin_unlock(&dentry->d_lock); 970 spin_unlock(&dentry->d_lock);
972 dput(dentry); 971 dput(dentry);
973 } 972 }
974 } 973 }
975 } 974 }
976 975
977 /** 976 /**
978 * __mark_inode_dirty - internal function 977 * __mark_inode_dirty - internal function
979 * @inode: inode to mark 978 * @inode: inode to mark
980 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 979 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
981 * Mark an inode as dirty. Callers should use mark_inode_dirty or 980 * Mark an inode as dirty. Callers should use mark_inode_dirty or
982 * mark_inode_dirty_sync. 981 * mark_inode_dirty_sync.
983 * 982 *
984 * Put the inode on the super block's dirty list. 983 * Put the inode on the super block's dirty list.
985 * 984 *
986 * CAREFUL! We mark it dirty unconditionally, but move it onto the 985 * CAREFUL! We mark it dirty unconditionally, but move it onto the
987 * dirty list only if it is hashed or if it refers to a blockdev. 986 * dirty list only if it is hashed or if it refers to a blockdev.
988 * If it was not hashed, it will never be added to the dirty list 987 * If it was not hashed, it will never be added to the dirty list
989 * even if it is later hashed, as it will have been marked dirty already. 988 * even if it is later hashed, as it will have been marked dirty already.
990 * 989 *
991 * In short, make sure you hash any inodes _before_ you start marking 990 * In short, make sure you hash any inodes _before_ you start marking
992 * them dirty. 991 * them dirty.
993 * 992 *
994 * This function *must* be atomic for the I_DIRTY_PAGES case - 993 * This function *must* be atomic for the I_DIRTY_PAGES case -
995 * set_page_dirty() is called under spinlock in several places. 994 * set_page_dirty() is called under spinlock in several places.
996 * 995 *
997 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 996 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
998 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 997 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
999 * the kernel-internal blockdev inode represents the dirtying time of the 998 * the kernel-internal blockdev inode represents the dirtying time of the
1000 * blockdev's pages. This is why for I_DIRTY_PAGES we always use 999 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1001 * page->mapping->host, so the page-dirtying time is recorded in the internal 1000 * page->mapping->host, so the page-dirtying time is recorded in the internal
1002 * blockdev inode. 1001 * blockdev inode.
1003 */ 1002 */
1004 void __mark_inode_dirty(struct inode *inode, int flags) 1003 void __mark_inode_dirty(struct inode *inode, int flags)
1005 { 1004 {
1006 struct super_block *sb = inode->i_sb; 1005 struct super_block *sb = inode->i_sb;
1007 1006
1008 /* 1007 /*
1009 * Don't do this for I_DIRTY_PAGES - that doesn't actually 1008 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1010 * dirty the inode itself 1009 * dirty the inode itself
1011 */ 1010 */
1012 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 1011 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1013 if (sb->s_op->dirty_inode) 1012 if (sb->s_op->dirty_inode)
1014 sb->s_op->dirty_inode(inode); 1013 sb->s_op->dirty_inode(inode);
1015 } 1014 }
1016 1015
1017 /* 1016 /*
1018 * make sure that changes are seen by all cpus before we test i_state 1017 * make sure that changes are seen by all cpus before we test i_state
1019 * -- mikulas 1018 * -- mikulas
1020 */ 1019 */
1021 smp_mb(); 1020 smp_mb();
1022 1021
1023 /* avoid the locking if we can */ 1022 /* avoid the locking if we can */
1024 if ((inode->i_state & flags) == flags) 1023 if ((inode->i_state & flags) == flags)
1025 return; 1024 return;
1026 1025
1027 if (unlikely(block_dump)) 1026 if (unlikely(block_dump))
1028 block_dump___mark_inode_dirty(inode); 1027 block_dump___mark_inode_dirty(inode);
1029 1028
1030 spin_lock(&inode_lock); 1029 spin_lock(&inode_lock);
1031 if ((inode->i_state & flags) != flags) { 1030 if ((inode->i_state & flags) != flags) {
1032 const int was_dirty = inode->i_state & I_DIRTY; 1031 const int was_dirty = inode->i_state & I_DIRTY;
1033 1032
1034 inode->i_state |= flags; 1033 inode->i_state |= flags;
1035 1034
1036 /* 1035 /*
1037 * If the inode is being synced, just update its dirty state. 1036 * If the inode is being synced, just update its dirty state.
1038 * The unlocker will place the inode on the appropriate 1037 * The unlocker will place the inode on the appropriate
1039 * superblock list, based upon its state. 1038 * superblock list, based upon its state.
1040 */ 1039 */
1041 if (inode->i_state & I_SYNC) 1040 if (inode->i_state & I_SYNC)
1042 goto out; 1041 goto out;
1043 1042
1044 /* 1043 /*
1045 * Only add valid (hashed) inodes to the superblock's 1044 * Only add valid (hashed) inodes to the superblock's
1046 * dirty list. Add blockdev inodes as well. 1045 * dirty list. Add blockdev inodes as well.
1047 */ 1046 */
1048 if (!S_ISBLK(inode->i_mode)) { 1047 if (!S_ISBLK(inode->i_mode)) {
1049 if (hlist_unhashed(&inode->i_hash)) 1048 if (hlist_unhashed(&inode->i_hash))
1050 goto out; 1049 goto out;
1051 } 1050 }
1052 if (inode->i_state & (I_FREEING|I_CLEAR)) 1051 if (inode->i_state & (I_FREEING|I_CLEAR))
1053 goto out; 1052 goto out;
1054 1053
1055 /* 1054 /*
1056 * If the inode was already on b_dirty/b_io/b_more_io, don't 1055 * If the inode was already on b_dirty/b_io/b_more_io, don't
1057 * reposition it (that would break b_dirty time-ordering). 1056 * reposition it (that would break b_dirty time-ordering).
1058 */ 1057 */
1059 if (!was_dirty) { 1058 if (!was_dirty) {
1060 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 1059 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1061 struct backing_dev_info *bdi = wb->bdi; 1060 struct backing_dev_info *bdi = wb->bdi;
1062 1061
1063 if (bdi_cap_writeback_dirty(bdi) && 1062 if (bdi_cap_writeback_dirty(bdi) &&
1064 !test_bit(BDI_registered, &bdi->state)) { 1063 !test_bit(BDI_registered, &bdi->state)) {
1065 WARN_ON(1); 1064 WARN_ON(1);
1066 printk(KERN_ERR "bdi-%s not registered\n", 1065 printk(KERN_ERR "bdi-%s not registered\n",
1067 bdi->name); 1066 bdi->name);
1068 } 1067 }
1069 1068
1070 inode->dirtied_when = jiffies; 1069 inode->dirtied_when = jiffies;
1071 list_move(&inode->i_list, &wb->b_dirty); 1070 list_move(&inode->i_list, &wb->b_dirty);
1072 } 1071 }
1073 } 1072 }
1074 out: 1073 out:
1075 spin_unlock(&inode_lock); 1074 spin_unlock(&inode_lock);
1076 } 1075 }
1077 EXPORT_SYMBOL(__mark_inode_dirty); 1076 EXPORT_SYMBOL(__mark_inode_dirty);
1078 1077
1079 /* 1078 /*
1080 * Write out a superblock's list of dirty inodes. A wait will be performed 1079 * Write out a superblock's list of dirty inodes. A wait will be performed
1081 * upon no inodes, all inodes or the final one, depending upon sync_mode. 1080 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1082 * 1081 *
1083 * If older_than_this is non-NULL, then only write out inodes which 1082 * If older_than_this is non-NULL, then only write out inodes which
1084 * had their first dirtying at a time earlier than *older_than_this. 1083 * had their first dirtying at a time earlier than *older_than_this.
1085 * 1084 *
1086 * If we're a pdlfush thread, then implement pdflush collision avoidance 1085 * If we're a pdlfush thread, then implement pdflush collision avoidance
1087 * against the entire list. 1086 * against the entire list.
1088 * 1087 *
1089 * If `bdi' is non-zero then we're being asked to writeback a specific queue. 1088 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1090 * This function assumes that the blockdev superblock's inodes are backed by 1089 * This function assumes that the blockdev superblock's inodes are backed by
1091 * a variety of queues, so all inodes are searched. For other superblocks, 1090 * a variety of queues, so all inodes are searched. For other superblocks,
1092 * assume that all inodes are backed by the same queue. 1091 * assume that all inodes are backed by the same queue.
1093 * 1092 *
1094 * The inodes to be written are parked on bdi->b_io. They are moved back onto 1093 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1095 * bdi->b_dirty as they are selected for writing. This way, none can be missed 1094 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1096 * on the writer throttling path, and we get decent balancing between many 1095 * on the writer throttling path, and we get decent balancing between many
1097 * throttled threads: we don't want them all piling up on inode_sync_wait. 1096 * throttled threads: we don't want them all piling up on inode_sync_wait.
1098 */ 1097 */
1099 static void wait_sb_inodes(struct super_block *sb) 1098 static void wait_sb_inodes(struct super_block *sb)
1100 { 1099 {
1101 struct inode *inode, *old_inode = NULL; 1100 struct inode *inode, *old_inode = NULL;
1102 1101
1103 /* 1102 /*
1104 * We need to be protected against the filesystem going from 1103 * We need to be protected against the filesystem going from
1105 * r/o to r/w or vice versa. 1104 * r/o to r/w or vice versa.
1106 */ 1105 */
1107 WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1106 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1108 1107
1109 spin_lock(&inode_lock); 1108 spin_lock(&inode_lock);
1110 1109
1111 /* 1110 /*
1112 * Data integrity sync. Must wait for all pages under writeback, 1111 * Data integrity sync. Must wait for all pages under writeback,
1113 * because there may have been pages dirtied before our sync 1112 * because there may have been pages dirtied before our sync
1114 * call, but which had writeout started before we write it out. 1113 * call, but which had writeout started before we write it out.
1115 * In which case, the inode may not be on the dirty list, but 1114 * In which case, the inode may not be on the dirty list, but
1116 * we still have to wait for that writeout. 1115 * we still have to wait for that writeout.
1117 */ 1116 */
1118 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1117 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1119 struct address_space *mapping; 1118 struct address_space *mapping;
1120 1119
1121 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW)) 1120 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW))
1122 continue; 1121 continue;
1123 mapping = inode->i_mapping; 1122 mapping = inode->i_mapping;
1124 if (mapping->nrpages == 0) 1123 if (mapping->nrpages == 0)
1125 continue; 1124 continue;
1126 __iget(inode); 1125 __iget(inode);
1127 spin_unlock(&inode_lock); 1126 spin_unlock(&inode_lock);
1128 /* 1127 /*
1129 * We hold a reference to 'inode' so it couldn't have 1128 * We hold a reference to 'inode' so it couldn't have
1130 * been removed from s_inodes list while we dropped the 1129 * been removed from s_inodes list while we dropped the
1131 * inode_lock. We cannot iput the inode now as we can 1130 * inode_lock. We cannot iput the inode now as we can
1132 * be holding the last reference and we cannot iput it 1131 * be holding the last reference and we cannot iput it
1133 * under inode_lock. So we keep the reference and iput 1132 * under inode_lock. So we keep the reference and iput
1134 * it later. 1133 * it later.
1135 */ 1134 */
1136 iput(old_inode); 1135 iput(old_inode);
1137 old_inode = inode; 1136 old_inode = inode;
1138 1137
1139 filemap_fdatawait(mapping); 1138 filemap_fdatawait(mapping);
1140 1139
1141 cond_resched(); 1140 cond_resched();
1142 1141
1143 spin_lock(&inode_lock); 1142 spin_lock(&inode_lock);
1144 } 1143 }
1145 spin_unlock(&inode_lock); 1144 spin_unlock(&inode_lock);
1146 iput(old_inode); 1145 iput(old_inode);
1147 } 1146 }
1148 1147
1149 /** 1148 /**
1150 * writeback_inodes_sb - writeback dirty inodes from given super_block 1149 * writeback_inodes_sb - writeback dirty inodes from given super_block
1151 * @sb: the superblock 1150 * @sb: the superblock
1152 * 1151 *
1153 * Start writeback on some inodes on this super_block. No guarantees are made 1152 * Start writeback on some inodes on this super_block. No guarantees are made
1154 * on how many (if any) will be written, and this function does not wait 1153 * on how many (if any) will be written, and this function does not wait
1155 * for IO completion of submitted IO. The number of pages submitted is 1154 * for IO completion of submitted IO. The number of pages submitted is
1156 * returned. 1155 * returned.
1157 */ 1156 */
1158 void writeback_inodes_sb(struct super_block *sb) 1157 void writeback_inodes_sb(struct super_block *sb)
1159 { 1158 {
1160 unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); 1159 unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY);
1161 unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); 1160 unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS);
1162 long nr_to_write; 1161 long nr_to_write;
1163 1162
1164 nr_to_write = nr_dirty + nr_unstable + 1163 nr_to_write = nr_dirty + nr_unstable +
1165 (inodes_stat.nr_inodes - inodes_stat.nr_unused); 1164 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
1166 1165
1167 bdi_writeback_all(sb, nr_to_write); 1166 bdi_writeback_all(sb, nr_to_write);
1168 } 1167 }
1169 EXPORT_SYMBOL(writeback_inodes_sb); 1168 EXPORT_SYMBOL(writeback_inodes_sb);
1170 1169
1171 /** 1170 /**
1172 * sync_inodes_sb - sync sb inode pages 1171 * sync_inodes_sb - sync sb inode pages
1173 * @sb: the superblock 1172 * @sb: the superblock
1174 * 1173 *
1175 * This function writes and waits on any dirty inode belonging to this 1174 * This function writes and waits on any dirty inode belonging to this
1176 * super_block. The number of pages synced is returned. 1175 * super_block. The number of pages synced is returned.
1177 */ 1176 */
1178 void sync_inodes_sb(struct super_block *sb) 1177 void sync_inodes_sb(struct super_block *sb)
1179 { 1178 {
1180 bdi_sync_writeback(sb->s_bdi, sb); 1179 bdi_sync_writeback(sb->s_bdi, sb);
1181 wait_sb_inodes(sb); 1180 wait_sb_inodes(sb);
1182 } 1181 }
1183 EXPORT_SYMBOL(sync_inodes_sb); 1182 EXPORT_SYMBOL(sync_inodes_sb);
1184 1183
1185 /** 1184 /**
1186 * write_inode_now - write an inode to disk 1185 * write_inode_now - write an inode to disk
1187 * @inode: inode to write to disk 1186 * @inode: inode to write to disk
1188 * @sync: whether the write should be synchronous or not 1187 * @sync: whether the write should be synchronous or not
1189 * 1188 *
1190 * This function commits an inode to disk immediately if it is dirty. This is 1189 * This function commits an inode to disk immediately if it is dirty. This is
1191 * primarily needed by knfsd. 1190 * primarily needed by knfsd.
1192 * 1191 *
1193 * The caller must either have a ref on the inode or must have set I_WILL_FREE. 1192 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1194 */ 1193 */
1195 int write_inode_now(struct inode *inode, int sync) 1194 int write_inode_now(struct inode *inode, int sync)
1196 { 1195 {
1197 int ret; 1196 int ret;
1198 struct writeback_control wbc = { 1197 struct writeback_control wbc = {
1199 .nr_to_write = LONG_MAX, 1198 .nr_to_write = LONG_MAX,
1200 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1199 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1201 .range_start = 0, 1200 .range_start = 0,
1202 .range_end = LLONG_MAX, 1201 .range_end = LLONG_MAX,
1203 }; 1202 };
1204 1203
1205 if (!mapping_cap_writeback_dirty(inode->i_mapping)) 1204 if (!mapping_cap_writeback_dirty(inode->i_mapping))
1206 wbc.nr_to_write = 0; 1205 wbc.nr_to_write = 0;
1207 1206
1208 might_sleep(); 1207 might_sleep();
1209 spin_lock(&inode_lock); 1208 spin_lock(&inode_lock);
1210 ret = writeback_single_inode(inode, &wbc); 1209 ret = writeback_single_inode(inode, &wbc);
1211 spin_unlock(&inode_lock); 1210 spin_unlock(&inode_lock);
1212 if (sync) 1211 if (sync)
1213 inode_sync_wait(inode); 1212 inode_sync_wait(inode);
1214 return ret; 1213 return ret;
1215 } 1214 }
1216 EXPORT_SYMBOL(write_inode_now); 1215 EXPORT_SYMBOL(write_inode_now);
1217 1216
1218 /** 1217 /**
1219 * sync_inode - write an inode and its pages to disk. 1218 * sync_inode - write an inode and its pages to disk.
1220 * @inode: the inode to sync 1219 * @inode: the inode to sync
1221 * @wbc: controls the writeback mode 1220 * @wbc: controls the writeback mode
1222 * 1221 *
1223 * sync_inode() will write an inode and its pages to disk. It will also 1222 * sync_inode() will write an inode and its pages to disk. It will also
1224 * correctly update the inode on its superblock's dirty inode lists and will 1223 * correctly update the inode on its superblock's dirty inode lists and will
1225 * update inode->i_state. 1224 * update inode->i_state.
1226 * 1225 *
1227 * The caller must have a ref on the inode. 1226 * The caller must have a ref on the inode.
1228 */ 1227 */
1229 int sync_inode(struct inode *inode, struct writeback_control *wbc) 1228 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1230 { 1229 {
1231 int ret; 1230 int ret;
1232 1231
1233 spin_lock(&inode_lock); 1232 spin_lock(&inode_lock);