blob: 3b2f9cb829864753b810bd87e15e16cafc74215b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -070011#include <linux/percpu_counter.h>
12#include <linux/log2.h>
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -070013#include <linux/proportions.h>
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -070014#include <linux/kernel.h>
Miklos Szeredie4ad08f2008-04-30 00:54:37 -070015#include <linux/fs.h>
Jens Axboe03ba3782009-09-09 09:08:54 +020016#include <linux/sched.h>
Matthew Garrett31373d02010-04-06 14:25:14 +020017#include <linux/timer.h>
Jens Axboe03ba3782009-09-09 09:08:54 +020018#include <linux/writeback.h>
Arun Sharma600634972011-07-26 16:09:06 -070019#include <linux/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020
Andrew Morton3fcfab12006-10-19 23:28:16 -070021struct page;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -070022struct device;
Miklos Szeredi76f14182008-04-30 00:54:36 -070023struct dentry;
Andrew Morton3fcfab12006-10-19 23:28:16 -070024
Linus Torvalds1da177e2005-04-16 15:20:36 -070025/*
26 * Bits in backing_dev_info.state
27 */
28enum bdi_state {
Jens Axboe03ba3782009-09-09 09:08:54 +020029 BDI_pending, /* On its way to being activated */
30 BDI_wb_alloc, /* Default embedded wb allocated */
Jens Axboe1faa16d2009-04-06 14:48:01 +020031 BDI_async_congested, /* The async (write) queue is getting full */
32 BDI_sync_congested, /* The sync queue is getting full */
Jens Axboe500b0672009-09-09 09:10:25 +020033 BDI_registered, /* bdi_register() was done */
Jan Kara81d73a32010-08-11 14:17:44 -070034 BDI_writeback_running, /* Writeback is in progress */
Linus Torvalds1da177e2005-04-16 15:20:36 -070035 BDI_unused, /* Available bits start here */
36};
37
38typedef int (congested_fn)(void *, int);
39
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -070040enum bdi_stat_item {
Peter Zijlstrac9e51e42007-10-16 23:25:47 -070041 BDI_RECLAIMABLE,
Peter Zijlstra69cb51d2007-10-16 23:25:48 -070042 BDI_WRITEBACK,
Jan Karaf7d2b1e2010-12-08 22:44:24 -060043 BDI_WRITTEN,
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -070044 NR_BDI_STAT_ITEMS
45};
46
47#define BDI_STAT_BATCH (8*(1+ilog2(nr_cpu_ids)))
48
Jens Axboe03ba3782009-09-09 09:08:54 +020049struct bdi_writeback {
Artem Bityutskiyecd58402010-07-25 14:29:18 +030050 struct backing_dev_info *bdi; /* our parent bdi */
Jens Axboe03ba3782009-09-09 09:08:54 +020051 unsigned int nr;
52
Artem Bityutskiyecd58402010-07-25 14:29:18 +030053 unsigned long last_old_flush; /* last old data flush */
54 unsigned long last_active; /* last time bdi thread was active */
Jens Axboe03ba3782009-09-09 09:08:54 +020055
Artem Bityutskiyecd58402010-07-25 14:29:18 +030056 struct task_struct *task; /* writeback thread */
Artem Bityutskiy64677162010-07-25 14:29:22 +030057 struct timer_list wakeup_timer; /* used for delayed bdi thread wakeup */
Artem Bityutskiyecd58402010-07-25 14:29:18 +030058 struct list_head b_dirty; /* dirty inodes */
59 struct list_head b_io; /* parked for writeback */
60 struct list_head b_more_io; /* parked for more writeback */
Christoph Hellwigf758eea2011-04-21 18:19:44 -060061 spinlock_t list_lock; /* protects the b_* lists */
Jens Axboe03ba3782009-09-09 09:08:54 +020062};
63
Linus Torvalds1da177e2005-04-16 15:20:36 -070064struct backing_dev_info {
Jens Axboe66f3b8e2009-09-02 09:19:46 +020065 struct list_head bdi_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 unsigned long ra_pages; /* max readahead in PAGE_CACHE_SIZE units */
67 unsigned long state; /* Always use atomic bitops on this */
68 unsigned int capabilities; /* Device capabilities */
69 congested_fn *congested_fn; /* Function pointer if device is md/dm */
70 void *congested_data; /* Pointer to aux data for congested func */
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -070071
Jens Axboed9938312009-06-12 14:45:52 +020072 char *name;
73
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -070074 struct percpu_counter bdi_stat[NR_BDI_STAT_ITEMS];
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -070075
Wu Fengguange98be2d2010-08-29 11:22:30 -060076 unsigned long bw_time_stamp; /* last time write bw is updated */
77 unsigned long written_stamp; /* pages written at bw_time_stamp */
78 unsigned long write_bandwidth; /* the estimated write bandwidth */
79 unsigned long avg_write_bandwidth; /* further smoothed write bw */
80
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -070081 struct prop_local_percpu completions;
82 int dirty_exceeded;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -070083
Peter Zijlstra189d3c42008-04-30 00:54:35 -070084 unsigned int min_ratio;
Peter Zijlstraa42dde02008-04-30 00:54:36 -070085 unsigned int max_ratio, max_prop_frac;
Peter Zijlstra189d3c42008-04-30 00:54:35 -070086
Jens Axboe03ba3782009-09-09 09:08:54 +020087 struct bdi_writeback wb; /* default writeback info for this bdi */
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020088 spinlock_t wb_lock; /* protects work_list */
Miklos Szeredi76f14182008-04-30 00:54:36 -070089
Jens Axboe03ba3782009-09-09 09:08:54 +020090 struct list_head work_list;
91
92 struct device *dev;
Jens Axboe66f3b8e2009-09-02 09:19:46 +020093
Matthew Garrett31373d02010-04-06 14:25:14 +020094 struct timer_list laptop_mode_wb_timer;
95
Miklos Szeredi76f14182008-04-30 00:54:36 -070096#ifdef CONFIG_DEBUG_FS
97 struct dentry *debug_dir;
98 struct dentry *debug_stats;
99#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100};
101
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700102int bdi_init(struct backing_dev_info *bdi);
103void bdi_destroy(struct backing_dev_info *bdi);
104
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700105int bdi_register(struct backing_dev_info *bdi, struct device *parent,
106 const char *fmt, ...);
107int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev);
108void bdi_unregister(struct backing_dev_info *bdi);
Jens Axboec3c53202010-04-22 11:37:01 +0200109int bdi_setup_and_register(struct backing_dev_info *, char *, unsigned int);
Christoph Hellwigc5444192010-06-08 18:15:15 +0200110void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages);
111void bdi_start_background_writeback(struct backing_dev_info *bdi);
Christoph Hellwig08243902010-06-19 23:08:22 +0200112int bdi_writeback_thread(void *data);
Jens Axboe03ba3782009-09-09 09:08:54 +0200113int bdi_has_dirty_io(struct backing_dev_info *bdi);
Jens Axboe64231042010-05-21 20:00:35 +0200114void bdi_arm_supers_timer(void);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300115void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600116void bdi_lock_two(struct bdi_writeback *wb1, struct bdi_writeback *wb2);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700117
Jens Axboe03ba3782009-09-09 09:08:54 +0200118extern spinlock_t bdi_lock;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200119extern struct list_head bdi_list;
Namhyung Kim92c09c02010-10-26 14:22:03 -0700120extern struct list_head bdi_pending_list;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200121
Jens Axboe03ba3782009-09-09 09:08:54 +0200122static inline int wb_has_dirty_io(struct bdi_writeback *wb)
123{
124 return !list_empty(&wb->b_dirty) ||
125 !list_empty(&wb->b_io) ||
126 !list_empty(&wb->b_more_io);
127}
128
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700129static inline void __add_bdi_stat(struct backing_dev_info *bdi,
130 enum bdi_stat_item item, s64 amount)
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700131{
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700132 __percpu_counter_add(&bdi->bdi_stat[item], amount, BDI_STAT_BATCH);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700133}
134
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700135static inline void __inc_bdi_stat(struct backing_dev_info *bdi,
136 enum bdi_stat_item item)
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700137{
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700138 __add_bdi_stat(bdi, item, 1);
139}
140
141static inline void inc_bdi_stat(struct backing_dev_info *bdi,
142 enum bdi_stat_item item)
143{
144 unsigned long flags;
145
146 local_irq_save(flags);
147 __inc_bdi_stat(bdi, item);
148 local_irq_restore(flags);
149}
150
151static inline void __dec_bdi_stat(struct backing_dev_info *bdi,
152 enum bdi_stat_item item)
153{
154 __add_bdi_stat(bdi, item, -1);
155}
156
157static inline void dec_bdi_stat(struct backing_dev_info *bdi,
158 enum bdi_stat_item item)
159{
160 unsigned long flags;
161
162 local_irq_save(flags);
163 __dec_bdi_stat(bdi, item);
164 local_irq_restore(flags);
165}
166
167static inline s64 bdi_stat(struct backing_dev_info *bdi,
168 enum bdi_stat_item item)
169{
170 return percpu_counter_read_positive(&bdi->bdi_stat[item]);
171}
172
173static inline s64 __bdi_stat_sum(struct backing_dev_info *bdi,
174 enum bdi_stat_item item)
175{
176 return percpu_counter_sum_positive(&bdi->bdi_stat[item]);
177}
178
179static inline s64 bdi_stat_sum(struct backing_dev_info *bdi,
180 enum bdi_stat_item item)
181{
182 s64 sum;
183 unsigned long flags;
184
185 local_irq_save(flags);
186 sum = __bdi_stat_sum(bdi, item);
187 local_irq_restore(flags);
188
189 return sum;
190}
191
Miklos Szeredidd5656e2008-04-30 00:54:37 -0700192extern void bdi_writeout_inc(struct backing_dev_info *bdi);
193
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700194/*
195 * maximal error of a stat counter.
196 */
197static inline unsigned long bdi_stat_error(struct backing_dev_info *bdi)
198{
199#ifdef CONFIG_SMP
200 return nr_cpu_ids * BDI_STAT_BATCH;
201#else
202 return 1;
203#endif
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700204}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700206int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700207int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700208
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209/*
210 * Flags in backing_dev_info::capability
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700211 *
212 * The first three flags control whether dirty pages will contribute to the
213 * VM's accounting and whether writepages() should be called for dirty pages
214 * (something that would not, for example, be appropriate for ramfs)
215 *
216 * WARNING: these flags are closely related and should not normally be
217 * used separately. The BDI_CAP_NO_ACCT_AND_WRITEBACK combines these
218 * three flags into a single convenience macro.
219 *
220 * BDI_CAP_NO_ACCT_DIRTY: Dirty pages shouldn't contribute to accounting
221 * BDI_CAP_NO_WRITEBACK: Don't write pages back
222 * BDI_CAP_NO_ACCT_WB: Don't automatically account writeback pages
223 *
224 * These flags let !MMU mmap() govern direct device mapping vs immediate
225 * copying more easily for MAP_PRIVATE, especially for ROM filesystems.
226 *
227 * BDI_CAP_MAP_COPY: Copy can be mapped (MAP_PRIVATE)
228 * BDI_CAP_MAP_DIRECT: Can be mapped directly (MAP_SHARED)
229 * BDI_CAP_READ_MAP: Can be mapped for reading
230 * BDI_CAP_WRITE_MAP: Can be mapped for writing
231 * BDI_CAP_EXEC_MAP: Can be mapped for execution
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700232 *
233 * BDI_CAP_SWAP_BACKED: Count shmem/tmpfs objects as swap-backed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 */
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700235#define BDI_CAP_NO_ACCT_DIRTY 0x00000001
236#define BDI_CAP_NO_WRITEBACK 0x00000002
237#define BDI_CAP_MAP_COPY 0x00000004
238#define BDI_CAP_MAP_DIRECT 0x00000008
239#define BDI_CAP_READ_MAP 0x00000010
240#define BDI_CAP_WRITE_MAP 0x00000020
241#define BDI_CAP_EXEC_MAP 0x00000040
242#define BDI_CAP_NO_ACCT_WB 0x00000080
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700243#define BDI_CAP_SWAP_BACKED 0x00000100
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700244
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245#define BDI_CAP_VMFLAGS \
246 (BDI_CAP_READ_MAP | BDI_CAP_WRITE_MAP | BDI_CAP_EXEC_MAP)
247
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700248#define BDI_CAP_NO_ACCT_AND_WRITEBACK \
249 (BDI_CAP_NO_WRITEBACK | BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_ACCT_WB)
250
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251#if defined(VM_MAYREAD) && \
252 (BDI_CAP_READ_MAP != VM_MAYREAD || \
253 BDI_CAP_WRITE_MAP != VM_MAYWRITE || \
254 BDI_CAP_EXEC_MAP != VM_MAYEXEC)
255#error please change backing_dev_info::capabilities flags
256#endif
257
258extern struct backing_dev_info default_backing_dev_info;
Jörn Engel5129a462010-04-25 08:54:42 +0200259extern struct backing_dev_info noop_backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261int writeback_in_progress(struct backing_dev_info *bdi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262
263static inline int bdi_congested(struct backing_dev_info *bdi, int bdi_bits)
264{
265 if (bdi->congested_fn)
266 return bdi->congested_fn(bdi->congested_data, bdi_bits);
267 return (bdi->state & bdi_bits);
268}
269
270static inline int bdi_read_congested(struct backing_dev_info *bdi)
271{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200272 return bdi_congested(bdi, 1 << BDI_sync_congested);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273}
274
275static inline int bdi_write_congested(struct backing_dev_info *bdi)
276{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200277 return bdi_congested(bdi, 1 << BDI_async_congested);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278}
279
280static inline int bdi_rw_congested(struct backing_dev_info *bdi)
281{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200282 return bdi_congested(bdi, (1 << BDI_sync_congested) |
283 (1 << BDI_async_congested));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284}
285
Trond Myklebust373c0a72009-07-11 10:06:54 -0400286enum {
287 BLK_RW_ASYNC = 0,
288 BLK_RW_SYNC = 1,
289};
290
Jens Axboe8aa7e842009-07-09 14:52:32 +0200291void clear_bdi_congested(struct backing_dev_info *bdi, int sync);
292void set_bdi_congested(struct backing_dev_info *bdi, int sync);
293long congestion_wait(int sync, long timeout);
Mel Gorman0e093d992010-10-26 14:21:45 -0700294long wait_iff_congested(struct zone *zone, int sync, long timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700296static inline bool bdi_cap_writeback_dirty(struct backing_dev_info *bdi)
297{
298 return !(bdi->capabilities & BDI_CAP_NO_WRITEBACK);
299}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700301static inline bool bdi_cap_account_dirty(struct backing_dev_info *bdi)
302{
303 return !(bdi->capabilities & BDI_CAP_NO_ACCT_DIRTY);
304}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700306static inline bool bdi_cap_account_writeback(struct backing_dev_info *bdi)
307{
308 /* Paranoia: BDI_CAP_NO_WRITEBACK implies BDI_CAP_NO_ACCT_WB */
309 return !(bdi->capabilities & (BDI_CAP_NO_ACCT_WB |
310 BDI_CAP_NO_WRITEBACK));
311}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700313static inline bool bdi_cap_swap_backed(struct backing_dev_info *bdi)
314{
315 return bdi->capabilities & BDI_CAP_SWAP_BACKED;
316}
317
Jens Axboe03ba3782009-09-09 09:08:54 +0200318static inline bool bdi_cap_flush_forker(struct backing_dev_info *bdi)
319{
320 return bdi == &default_backing_dev_info;
321}
322
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700323static inline bool mapping_cap_writeback_dirty(struct address_space *mapping)
324{
325 return bdi_cap_writeback_dirty(mapping->backing_dev_info);
326}
327
328static inline bool mapping_cap_account_dirty(struct address_space *mapping)
329{
330 return bdi_cap_account_dirty(mapping->backing_dev_info);
331}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700333static inline bool mapping_cap_swap_backed(struct address_space *mapping)
334{
335 return bdi_cap_swap_backed(mapping->backing_dev_info);
336}
337
Jens Axboe03ba3782009-09-09 09:08:54 +0200338static inline int bdi_sched_wait(void *word)
339{
340 schedule();
341 return 0;
342}
343
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344#endif /* _LINUX_BACKING_DEV_H */