Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * include/linux/backing-dev.h | |
3 | * | |
4 | * low-level device information and state which is propagated up through | |
5 | * to high-level code. | |
6 | */ | |
7 | ||
8 | #ifndef _LINUX_BACKING_DEV_H | |
9 | #define _LINUX_BACKING_DEV_H | |
10 | ||
b2e8fb6e PZ |
11 | #include <linux/percpu_counter.h> |
12 | #include <linux/log2.h> | |
eb608e3a | 13 | #include <linux/flex_proportions.h> |
cf0ca9fe | 14 | #include <linux/kernel.h> |
e4ad08fe | 15 | #include <linux/fs.h> |
03ba3782 | 16 | #include <linux/sched.h> |
31373d09 | 17 | #include <linux/timer.h> |
03ba3782 | 18 | #include <linux/writeback.h> |
60063497 | 19 | #include <linux/atomic.h> |
3965c9ae | 20 | #include <linux/sysctl.h> |
839a8e86 | 21 | #include <linux/workqueue.h> |
1da177e4 | 22 | |
3fcfab16 | 23 | struct page; |
cf0ca9fe | 24 | struct device; |
76f1418b | 25 | struct dentry; |
3fcfab16 | 26 | |
1da177e4 LT |
27 | /* |
28 | * Bits in backing_dev_info.state | |
29 | */ | |
30 | enum bdi_state { | |
1faa16d2 JA |
31 | BDI_async_congested, /* The async (write) queue is getting full */ |
32 | BDI_sync_congested, /* The sync queue is getting full */ | |
500b067c | 33 | BDI_registered, /* bdi_register() was done */ |
81d73a32 | 34 | BDI_writeback_running, /* Writeback is in progress */ |
1da177e4 LT |
35 | }; |
36 | ||
37 | typedef int (congested_fn)(void *, int); | |
38 | ||
b2e8fb6e | 39 | enum bdi_stat_item { |
c9e51e41 | 40 | BDI_RECLAIMABLE, |
69cb51d1 | 41 | BDI_WRITEBACK, |
c8e28ce0 | 42 | BDI_DIRTIED, |
f7d2b1ec | 43 | BDI_WRITTEN, |
b2e8fb6e PZ |
44 | NR_BDI_STAT_ITEMS |
45 | }; | |
46 | ||
47 | #define BDI_STAT_BATCH (8*(1+ilog2(nr_cpu_ids))) | |
48 | ||
03ba3782 | 49 | struct bdi_writeback { |
ecd58403 | 50 | struct backing_dev_info *bdi; /* our parent bdi */ |
03ba3782 | 51 | |
ecd58403 | 52 | unsigned long last_old_flush; /* last old data flush */ |
03ba3782 | 53 | |
839a8e86 | 54 | struct delayed_work dwork; /* work item used for writeback */ |
ecd58403 AB |
55 | struct list_head b_dirty; /* dirty inodes */ |
56 | struct list_head b_io; /* parked for writeback */ | |
57 | struct list_head b_more_io; /* parked for more writeback */ | |
0ae45f63 | 58 | struct list_head b_dirty_time; /* time stamps are dirty */ |
f758eeab | 59 | spinlock_t list_lock; /* protects the b_* lists */ |
03ba3782 JA |
60 | }; |
61 | ||
1da177e4 | 62 | struct backing_dev_info { |
66f3b8e2 | 63 | struct list_head bdi_list; |
1da177e4 LT |
64 | unsigned long ra_pages; /* max readahead in PAGE_CACHE_SIZE units */ |
65 | unsigned long state; /* Always use atomic bitops on this */ | |
66 | unsigned int capabilities; /* Device capabilities */ | |
67 | congested_fn *congested_fn; /* Function pointer if device is md/dm */ | |
68 | void *congested_data; /* Pointer to aux data for congested func */ | |
b2e8fb6e | 69 | |
d993831f JA |
70 | char *name; |
71 | ||
b2e8fb6e | 72 | struct percpu_counter bdi_stat[NR_BDI_STAT_ITEMS]; |
04fbfdc1 | 73 | |
e98be2d5 | 74 | unsigned long bw_time_stamp; /* last time write bw is updated */ |
be3ffa27 | 75 | unsigned long dirtied_stamp; |
e98be2d5 WF |
76 | unsigned long written_stamp; /* pages written at bw_time_stamp */ |
77 | unsigned long write_bandwidth; /* the estimated write bandwidth */ | |
78 | unsigned long avg_write_bandwidth; /* further smoothed write bw */ | |
79 | ||
be3ffa27 WF |
80 | /* |
81 | * The base dirty throttle rate, re-calculated on every 200ms. | |
82 | * All the bdi tasks' dirty rate will be curbed under it. | |
7381131c WF |
83 | * @dirty_ratelimit tracks the estimated @balanced_dirty_ratelimit |
84 | * in small steps and is much more smooth/stable than the latter. | |
be3ffa27 WF |
85 | */ |
86 | unsigned long dirty_ratelimit; | |
7381131c | 87 | unsigned long balanced_dirty_ratelimit; |
be3ffa27 | 88 | |
eb608e3a | 89 | struct fprop_local_percpu completions; |
04fbfdc1 | 90 | int dirty_exceeded; |
cf0ca9fe | 91 | |
189d3c4a | 92 | unsigned int min_ratio; |
a42dde04 | 93 | unsigned int max_ratio, max_prop_frac; |
189d3c4a | 94 | |
03ba3782 | 95 | struct bdi_writeback wb; /* default writeback info for this bdi */ |
5acda9d1 | 96 | spinlock_t wb_lock; /* protects work_list & wb.dwork scheduling */ |
76f1418b | 97 | |
03ba3782 JA |
98 | struct list_head work_list; |
99 | ||
100 | struct device *dev; | |
66f3b8e2 | 101 | |
31373d09 MG |
102 | struct timer_list laptop_mode_wb_timer; |
103 | ||
76f1418b MS |
104 | #ifdef CONFIG_DEBUG_FS |
105 | struct dentry *debug_dir; | |
106 | struct dentry *debug_stats; | |
107 | #endif | |
1da177e4 LT |
108 | }; |
109 | ||
de1414a6 CH |
110 | struct backing_dev_info *inode_to_bdi(struct inode *inode); |
111 | ||
8077c0d9 | 112 | int __must_check bdi_init(struct backing_dev_info *bdi); |
b2e8fb6e PZ |
113 | void bdi_destroy(struct backing_dev_info *bdi); |
114 | ||
d2cc4dde | 115 | __printf(3, 4) |
cf0ca9fe PZ |
116 | int bdi_register(struct backing_dev_info *bdi, struct device *parent, |
117 | const char *fmt, ...); | |
118 | int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev); | |
119 | void bdi_unregister(struct backing_dev_info *bdi); | |
b4caecd4 | 120 | int __must_check bdi_setup_and_register(struct backing_dev_info *, char *); |
0e175a18 CW |
121 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
122 | enum wb_reason reason); | |
c5444198 | 123 | void bdi_start_background_writeback(struct backing_dev_info *bdi); |
839a8e86 | 124 | void bdi_writeback_workfn(struct work_struct *work); |
03ba3782 | 125 | int bdi_has_dirty_io(struct backing_dev_info *bdi); |
6467716a | 126 | void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi); |
cf0ca9fe | 127 | |
03ba3782 | 128 | extern spinlock_t bdi_lock; |
66f3b8e2 JA |
129 | extern struct list_head bdi_list; |
130 | ||
839a8e86 TH |
131 | extern struct workqueue_struct *bdi_wq; |
132 | ||
03ba3782 JA |
133 | static inline int wb_has_dirty_io(struct bdi_writeback *wb) |
134 | { | |
135 | return !list_empty(&wb->b_dirty) || | |
136 | !list_empty(&wb->b_io) || | |
137 | !list_empty(&wb->b_more_io); | |
138 | } | |
139 | ||
b2e8fb6e PZ |
140 | static inline void __add_bdi_stat(struct backing_dev_info *bdi, |
141 | enum bdi_stat_item item, s64 amount) | |
142 | { | |
143 | __percpu_counter_add(&bdi->bdi_stat[item], amount, BDI_STAT_BATCH); | |
144 | } | |
145 | ||
146 | static inline void __inc_bdi_stat(struct backing_dev_info *bdi, | |
147 | enum bdi_stat_item item) | |
148 | { | |
149 | __add_bdi_stat(bdi, item, 1); | |
150 | } | |
151 | ||
152 | static inline void inc_bdi_stat(struct backing_dev_info *bdi, | |
153 | enum bdi_stat_item item) | |
154 | { | |
155 | unsigned long flags; | |
156 | ||
157 | local_irq_save(flags); | |
158 | __inc_bdi_stat(bdi, item); | |
159 | local_irq_restore(flags); | |
160 | } | |
161 | ||
162 | static inline void __dec_bdi_stat(struct backing_dev_info *bdi, | |
163 | enum bdi_stat_item item) | |
164 | { | |
165 | __add_bdi_stat(bdi, item, -1); | |
166 | } | |
167 | ||
168 | static inline void dec_bdi_stat(struct backing_dev_info *bdi, | |
169 | enum bdi_stat_item item) | |
170 | { | |
171 | unsigned long flags; | |
172 | ||
173 | local_irq_save(flags); | |
174 | __dec_bdi_stat(bdi, item); | |
175 | local_irq_restore(flags); | |
176 | } | |
177 | ||
178 | static inline s64 bdi_stat(struct backing_dev_info *bdi, | |
179 | enum bdi_stat_item item) | |
180 | { | |
181 | return percpu_counter_read_positive(&bdi->bdi_stat[item]); | |
182 | } | |
183 | ||
184 | static inline s64 __bdi_stat_sum(struct backing_dev_info *bdi, | |
185 | enum bdi_stat_item item) | |
186 | { | |
187 | return percpu_counter_sum_positive(&bdi->bdi_stat[item]); | |
188 | } | |
189 | ||
190 | static inline s64 bdi_stat_sum(struct backing_dev_info *bdi, | |
191 | enum bdi_stat_item item) | |
e0bf68dd | 192 | { |
b2e8fb6e PZ |
193 | s64 sum; |
194 | unsigned long flags; | |
195 | ||
196 | local_irq_save(flags); | |
197 | sum = __bdi_stat_sum(bdi, item); | |
198 | local_irq_restore(flags); | |
199 | ||
200 | return sum; | |
e0bf68dd PZ |
201 | } |
202 | ||
dd5656e5 MS |
203 | extern void bdi_writeout_inc(struct backing_dev_info *bdi); |
204 | ||
b2e8fb6e PZ |
205 | /* |
206 | * maximal error of a stat counter. | |
207 | */ | |
208 | static inline unsigned long bdi_stat_error(struct backing_dev_info *bdi) | |
e0bf68dd | 209 | { |
b2e8fb6e PZ |
210 | #ifdef CONFIG_SMP |
211 | return nr_cpu_ids * BDI_STAT_BATCH; | |
212 | #else | |
213 | return 1; | |
214 | #endif | |
e0bf68dd | 215 | } |
1da177e4 | 216 | |
189d3c4a | 217 | int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio); |
a42dde04 | 218 | int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio); |
189d3c4a | 219 | |
1da177e4 LT |
220 | /* |
221 | * Flags in backing_dev_info::capability | |
e4ad08fe MS |
222 | * |
223 | * The first three flags control whether dirty pages will contribute to the | |
224 | * VM's accounting and whether writepages() should be called for dirty pages | |
225 | * (something that would not, for example, be appropriate for ramfs) | |
226 | * | |
227 | * WARNING: these flags are closely related and should not normally be | |
228 | * used separately. The BDI_CAP_NO_ACCT_AND_WRITEBACK combines these | |
229 | * three flags into a single convenience macro. | |
230 | * | |
231 | * BDI_CAP_NO_ACCT_DIRTY: Dirty pages shouldn't contribute to accounting | |
232 | * BDI_CAP_NO_WRITEBACK: Don't write pages back | |
233 | * BDI_CAP_NO_ACCT_WB: Don't automatically account writeback pages | |
5a537485 | 234 | * BDI_CAP_STRICTLIMIT: Keep number of dirty pages below bdi threshold. |
1da177e4 | 235 | */ |
e4ad08fe MS |
236 | #define BDI_CAP_NO_ACCT_DIRTY 0x00000001 |
237 | #define BDI_CAP_NO_WRITEBACK 0x00000002 | |
b4caecd4 CH |
238 | #define BDI_CAP_NO_ACCT_WB 0x00000004 |
239 | #define BDI_CAP_STABLE_WRITES 0x00000008 | |
240 | #define BDI_CAP_STRICTLIMIT 0x00000010 | |
1da177e4 | 241 | |
e4ad08fe MS |
242 | #define BDI_CAP_NO_ACCT_AND_WRITEBACK \ |
243 | (BDI_CAP_NO_WRITEBACK | BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_ACCT_WB) | |
244 | ||
5129a469 | 245 | extern struct backing_dev_info noop_backing_dev_info; |
1da177e4 | 246 | |
1da177e4 | 247 | int writeback_in_progress(struct backing_dev_info *bdi); |
1da177e4 LT |
248 | |
249 | static inline int bdi_congested(struct backing_dev_info *bdi, int bdi_bits) | |
250 | { | |
251 | if (bdi->congested_fn) | |
252 | return bdi->congested_fn(bdi->congested_data, bdi_bits); | |
253 | return (bdi->state & bdi_bits); | |
254 | } | |
255 | ||
256 | static inline int bdi_read_congested(struct backing_dev_info *bdi) | |
257 | { | |
1faa16d2 | 258 | return bdi_congested(bdi, 1 << BDI_sync_congested); |
1da177e4 LT |
259 | } |
260 | ||
261 | static inline int bdi_write_congested(struct backing_dev_info *bdi) | |
262 | { | |
1faa16d2 | 263 | return bdi_congested(bdi, 1 << BDI_async_congested); |
1da177e4 LT |
264 | } |
265 | ||
266 | static inline int bdi_rw_congested(struct backing_dev_info *bdi) | |
267 | { | |
1faa16d2 JA |
268 | return bdi_congested(bdi, (1 << BDI_sync_congested) | |
269 | (1 << BDI_async_congested)); | |
1da177e4 LT |
270 | } |
271 | ||
373c0a7e TM |
272 | enum { |
273 | BLK_RW_ASYNC = 0, | |
274 | BLK_RW_SYNC = 1, | |
275 | }; | |
276 | ||
8aa7e847 JA |
277 | void clear_bdi_congested(struct backing_dev_info *bdi, int sync); |
278 | void set_bdi_congested(struct backing_dev_info *bdi, int sync); | |
279 | long congestion_wait(int sync, long timeout); | |
0e093d99 | 280 | long wait_iff_congested(struct zone *zone, int sync, long timeout); |
3965c9ae WL |
281 | int pdflush_proc_obsolete(struct ctl_table *table, int write, |
282 | void __user *buffer, size_t *lenp, loff_t *ppos); | |
1da177e4 | 283 | |
7d311cda DW |
284 | static inline bool bdi_cap_stable_pages_required(struct backing_dev_info *bdi) |
285 | { | |
286 | return bdi->capabilities & BDI_CAP_STABLE_WRITES; | |
287 | } | |
288 | ||
e4ad08fe MS |
289 | static inline bool bdi_cap_writeback_dirty(struct backing_dev_info *bdi) |
290 | { | |
291 | return !(bdi->capabilities & BDI_CAP_NO_WRITEBACK); | |
292 | } | |
293 | ||
294 | static inline bool bdi_cap_account_dirty(struct backing_dev_info *bdi) | |
295 | { | |
296 | return !(bdi->capabilities & BDI_CAP_NO_ACCT_DIRTY); | |
297 | } | |
1da177e4 | 298 | |
e4ad08fe MS |
299 | static inline bool bdi_cap_account_writeback(struct backing_dev_info *bdi) |
300 | { | |
301 | /* Paranoia: BDI_CAP_NO_WRITEBACK implies BDI_CAP_NO_ACCT_WB */ | |
302 | return !(bdi->capabilities & (BDI_CAP_NO_ACCT_WB | | |
303 | BDI_CAP_NO_WRITEBACK)); | |
304 | } | |
1da177e4 | 305 | |
e4ad08fe MS |
306 | static inline bool mapping_cap_writeback_dirty(struct address_space *mapping) |
307 | { | |
de1414a6 | 308 | return bdi_cap_writeback_dirty(inode_to_bdi(mapping->host)); |
e4ad08fe | 309 | } |
1da177e4 | 310 | |
e4ad08fe MS |
311 | static inline bool mapping_cap_account_dirty(struct address_space *mapping) |
312 | { | |
de1414a6 | 313 | return bdi_cap_account_dirty(inode_to_bdi(mapping->host)); |
e4ad08fe | 314 | } |
1da177e4 | 315 | |
03ba3782 JA |
316 | static inline int bdi_sched_wait(void *word) |
317 | { | |
318 | schedule(); | |
319 | return 0; | |
320 | } | |
321 | ||
1da177e4 | 322 | #endif /* _LINUX_BACKING_DEV_H */ |