blob: e488e9459a93f5e3e39ff339a7b0a1bbdee08165 [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>
Jan Karaeb608e32012-05-24 18:59:11 +020013#include <linux/flex_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>
Wanpeng Li3965c9a2012-07-31 16:41:52 -070020#include <linux/sysctl.h>
Tejun Heo839a8e82013-04-01 19:08:06 -070021#include <linux/workqueue.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022
Andrew Morton3fcfab12006-10-19 23:28:16 -070023struct page;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -070024struct device;
Miklos Szeredi76f14182008-04-30 00:54:36 -070025struct dentry;
Andrew Morton3fcfab12006-10-19 23:28:16 -070026
Linus Torvalds1da177e2005-04-16 15:20:36 -070027/*
28 * Bits in backing_dev_info.state
29 */
30enum bdi_state {
Jens Axboe03ba3782009-09-09 09:08:54 +020031 BDI_wb_alloc, /* Default embedded wb allocated */
Jens Axboe1faa16d2009-04-06 14:48:01 +020032 BDI_async_congested, /* The async (write) queue is getting full */
33 BDI_sync_congested, /* The sync queue is getting full */
Jens Axboe500b0672009-09-09 09:10:25 +020034 BDI_registered, /* bdi_register() was done */
Jan Kara81d73a32010-08-11 14:17:44 -070035 BDI_writeback_running, /* Writeback is in progress */
Linus Torvalds1da177e2005-04-16 15:20:36 -070036 BDI_unused, /* Available bits start here */
37};
38
39typedef int (congested_fn)(void *, int);
40
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -070041enum bdi_stat_item {
Peter Zijlstrac9e51e42007-10-16 23:25:47 -070042 BDI_RECLAIMABLE,
Peter Zijlstra69cb51d2007-10-16 23:25:48 -070043 BDI_WRITEBACK,
Wu Fengguangc8e28ce2011-01-23 10:07:47 -060044 BDI_DIRTIED,
Jan Karaf7d2b1e2010-12-08 22:44:24 -060045 BDI_WRITTEN,
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -070046 NR_BDI_STAT_ITEMS
47};
48
49#define BDI_STAT_BATCH (8*(1+ilog2(nr_cpu_ids)))
50
Jens Axboe03ba3782009-09-09 09:08:54 +020051struct bdi_writeback {
Artem Bityutskiyecd58402010-07-25 14:29:18 +030052 struct backing_dev_info *bdi; /* our parent bdi */
Jens Axboe03ba3782009-09-09 09:08:54 +020053 unsigned int nr;
54
Artem Bityutskiyecd58402010-07-25 14:29:18 +030055 unsigned long last_old_flush; /* last old data flush */
Jens Axboe03ba3782009-09-09 09:08:54 +020056
Tejun Heo839a8e82013-04-01 19:08:06 -070057 struct delayed_work dwork; /* work item used for writeback */
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 */
Wu Fengguangbe3ffa22011-06-12 10:51:31 -060077 unsigned long dirtied_stamp;
Wu Fengguange98be2d2010-08-29 11:22:30 -060078 unsigned long written_stamp; /* pages written at bw_time_stamp */
79 unsigned long write_bandwidth; /* the estimated write bandwidth */
80 unsigned long avg_write_bandwidth; /* further smoothed write bw */
81
Wu Fengguangbe3ffa22011-06-12 10:51:31 -060082 /*
83 * The base dirty throttle rate, re-calculated on every 200ms.
84 * All the bdi tasks' dirty rate will be curbed under it.
Wu Fengguang73811312011-08-26 15:53:24 -060085 * @dirty_ratelimit tracks the estimated @balanced_dirty_ratelimit
86 * in small steps and is much more smooth/stable than the latter.
Wu Fengguangbe3ffa22011-06-12 10:51:31 -060087 */
88 unsigned long dirty_ratelimit;
Wu Fengguang73811312011-08-26 15:53:24 -060089 unsigned long balanced_dirty_ratelimit;
Wu Fengguangbe3ffa22011-06-12 10:51:31 -060090
Jan Karaeb608e32012-05-24 18:59:11 +020091 struct fprop_local_percpu completions;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -070092 int dirty_exceeded;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -070093
Peter Zijlstra189d3c42008-04-30 00:54:35 -070094 unsigned int min_ratio;
Peter Zijlstraa42dde02008-04-30 00:54:36 -070095 unsigned int max_ratio, max_prop_frac;
Peter Zijlstra189d3c42008-04-30 00:54:35 -070096
Jens Axboe03ba3782009-09-09 09:08:54 +020097 struct bdi_writeback wb; /* default writeback info for this bdi */
Jan Kara5acda9d2014-04-03 14:46:23 -070098 spinlock_t wb_lock; /* protects work_list & wb.dwork scheduling */
Miklos Szeredi76f14182008-04-30 00:54:36 -070099
Jens Axboe03ba3782009-09-09 09:08:54 +0200100 struct list_head work_list;
101
102 struct device *dev;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200103
Matthew Garrett31373d02010-04-06 14:25:14 +0200104 struct timer_list laptop_mode_wb_timer;
105
Miklos Szeredi76f14182008-04-30 00:54:36 -0700106#ifdef CONFIG_DEBUG_FS
107 struct dentry *debug_dir;
108 struct dentry *debug_stats;
109#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110};
111
Mikulas Patocka8077c0d2013-10-14 12:14:13 -0400112int __must_check bdi_init(struct backing_dev_info *bdi);
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700113void bdi_destroy(struct backing_dev_info *bdi);
114
Joe Perchesd2cc4dd2012-11-29 08:37:03 -0600115__printf(3, 4)
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700116int bdi_register(struct backing_dev_info *bdi, struct device *parent,
117 const char *fmt, ...);
118int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev);
119void bdi_unregister(struct backing_dev_info *bdi);
Mikulas Patocka8077c0d2013-10-14 12:14:13 -0400120int __must_check bdi_setup_and_register(struct backing_dev_info *, char *, unsigned int);
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600121void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
122 enum wb_reason reason);
Christoph Hellwigc5444192010-06-08 18:15:15 +0200123void bdi_start_background_writeback(struct backing_dev_info *bdi);
Tejun Heo839a8e82013-04-01 19:08:06 -0700124void bdi_writeback_workfn(struct work_struct *work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200125int bdi_has_dirty_io(struct backing_dev_info *bdi);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300126void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600127void bdi_lock_two(struct bdi_writeback *wb1, struct bdi_writeback *wb2);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700128
Jens Axboe03ba3782009-09-09 09:08:54 +0200129extern spinlock_t bdi_lock;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200130extern struct list_head bdi_list;
131
Tejun Heo839a8e82013-04-01 19:08:06 -0700132extern struct workqueue_struct *bdi_wq;
133
Jens Axboe03ba3782009-09-09 09:08:54 +0200134static inline int wb_has_dirty_io(struct bdi_writeback *wb)
135{
136 return !list_empty(&wb->b_dirty) ||
137 !list_empty(&wb->b_io) ||
138 !list_empty(&wb->b_more_io);
139}
140
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700141static inline void __add_bdi_stat(struct backing_dev_info *bdi,
142 enum bdi_stat_item item, s64 amount)
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700143{
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700144 __percpu_counter_add(&bdi->bdi_stat[item], amount, BDI_STAT_BATCH);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700145}
146
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700147static inline void __inc_bdi_stat(struct backing_dev_info *bdi,
148 enum bdi_stat_item item)
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700149{
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700150 __add_bdi_stat(bdi, item, 1);
151}
152
153static inline void inc_bdi_stat(struct backing_dev_info *bdi,
154 enum bdi_stat_item item)
155{
156 unsigned long flags;
157
158 local_irq_save(flags);
159 __inc_bdi_stat(bdi, item);
160 local_irq_restore(flags);
161}
162
163static inline void __dec_bdi_stat(struct backing_dev_info *bdi,
164 enum bdi_stat_item item)
165{
166 __add_bdi_stat(bdi, item, -1);
167}
168
169static inline void dec_bdi_stat(struct backing_dev_info *bdi,
170 enum bdi_stat_item item)
171{
172 unsigned long flags;
173
174 local_irq_save(flags);
175 __dec_bdi_stat(bdi, item);
176 local_irq_restore(flags);
177}
178
179static inline s64 bdi_stat(struct backing_dev_info *bdi,
180 enum bdi_stat_item item)
181{
182 return percpu_counter_read_positive(&bdi->bdi_stat[item]);
183}
184
185static inline s64 __bdi_stat_sum(struct backing_dev_info *bdi,
186 enum bdi_stat_item item)
187{
188 return percpu_counter_sum_positive(&bdi->bdi_stat[item]);
189}
190
191static inline s64 bdi_stat_sum(struct backing_dev_info *bdi,
192 enum bdi_stat_item item)
193{
194 s64 sum;
195 unsigned long flags;
196
197 local_irq_save(flags);
198 sum = __bdi_stat_sum(bdi, item);
199 local_irq_restore(flags);
200
201 return sum;
202}
203
Miklos Szeredidd5656e2008-04-30 00:54:37 -0700204extern void bdi_writeout_inc(struct backing_dev_info *bdi);
205
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700206/*
207 * maximal error of a stat counter.
208 */
209static inline unsigned long bdi_stat_error(struct backing_dev_info *bdi)
210{
211#ifdef CONFIG_SMP
212 return nr_cpu_ids * BDI_STAT_BATCH;
213#else
214 return 1;
215#endif
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700216}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700218int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700219int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700220
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221/*
222 * Flags in backing_dev_info::capability
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700223 *
224 * The first three flags control whether dirty pages will contribute to the
225 * VM's accounting and whether writepages() should be called for dirty pages
226 * (something that would not, for example, be appropriate for ramfs)
227 *
228 * WARNING: these flags are closely related and should not normally be
229 * used separately. The BDI_CAP_NO_ACCT_AND_WRITEBACK combines these
230 * three flags into a single convenience macro.
231 *
232 * BDI_CAP_NO_ACCT_DIRTY: Dirty pages shouldn't contribute to accounting
233 * BDI_CAP_NO_WRITEBACK: Don't write pages back
234 * BDI_CAP_NO_ACCT_WB: Don't automatically account writeback pages
235 *
236 * These flags let !MMU mmap() govern direct device mapping vs immediate
237 * copying more easily for MAP_PRIVATE, especially for ROM filesystems.
238 *
239 * BDI_CAP_MAP_COPY: Copy can be mapped (MAP_PRIVATE)
240 * BDI_CAP_MAP_DIRECT: Can be mapped directly (MAP_SHARED)
241 * BDI_CAP_READ_MAP: Can be mapped for reading
242 * BDI_CAP_WRITE_MAP: Can be mapped for writing
243 * BDI_CAP_EXEC_MAP: Can be mapped for execution
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700244 *
245 * BDI_CAP_SWAP_BACKED: Count shmem/tmpfs objects as swap-backed.
Maxim Patlasov5a537482013-09-11 14:22:46 -0700246 *
247 * BDI_CAP_STRICTLIMIT: Keep number of dirty pages below bdi threshold.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 */
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700249#define BDI_CAP_NO_ACCT_DIRTY 0x00000001
250#define BDI_CAP_NO_WRITEBACK 0x00000002
251#define BDI_CAP_MAP_COPY 0x00000004
252#define BDI_CAP_MAP_DIRECT 0x00000008
253#define BDI_CAP_READ_MAP 0x00000010
254#define BDI_CAP_WRITE_MAP 0x00000020
255#define BDI_CAP_EXEC_MAP 0x00000040
256#define BDI_CAP_NO_ACCT_WB 0x00000080
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700257#define BDI_CAP_SWAP_BACKED 0x00000100
Darrick J. Wong7d311cd2013-02-21 16:42:48 -0800258#define BDI_CAP_STABLE_WRITES 0x00000200
Maxim Patlasov5a537482013-09-11 14:22:46 -0700259#define BDI_CAP_STRICTLIMIT 0x00000400
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700260
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261#define BDI_CAP_VMFLAGS \
262 (BDI_CAP_READ_MAP | BDI_CAP_WRITE_MAP | BDI_CAP_EXEC_MAP)
263
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700264#define BDI_CAP_NO_ACCT_AND_WRITEBACK \
265 (BDI_CAP_NO_WRITEBACK | BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_ACCT_WB)
266
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267#if defined(VM_MAYREAD) && \
268 (BDI_CAP_READ_MAP != VM_MAYREAD || \
269 BDI_CAP_WRITE_MAP != VM_MAYWRITE || \
270 BDI_CAP_EXEC_MAP != VM_MAYEXEC)
271#error please change backing_dev_info::capabilities flags
272#endif
273
274extern struct backing_dev_info default_backing_dev_info;
Jörn Engel5129a462010-04-25 08:54:42 +0200275extern struct backing_dev_info noop_backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277int writeback_in_progress(struct backing_dev_info *bdi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
279static inline int bdi_congested(struct backing_dev_info *bdi, int bdi_bits)
280{
281 if (bdi->congested_fn)
282 return bdi->congested_fn(bdi->congested_data, bdi_bits);
283 return (bdi->state & bdi_bits);
284}
285
286static inline int bdi_read_congested(struct backing_dev_info *bdi)
287{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200288 return bdi_congested(bdi, 1 << BDI_sync_congested);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289}
290
291static inline int bdi_write_congested(struct backing_dev_info *bdi)
292{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200293 return bdi_congested(bdi, 1 << BDI_async_congested);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294}
295
296static inline int bdi_rw_congested(struct backing_dev_info *bdi)
297{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200298 return bdi_congested(bdi, (1 << BDI_sync_congested) |
299 (1 << BDI_async_congested));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300}
301
Trond Myklebust373c0a72009-07-11 10:06:54 -0400302enum {
303 BLK_RW_ASYNC = 0,
304 BLK_RW_SYNC = 1,
305};
306
Jens Axboe8aa7e842009-07-09 14:52:32 +0200307void clear_bdi_congested(struct backing_dev_info *bdi, int sync);
308void set_bdi_congested(struct backing_dev_info *bdi, int sync);
309long congestion_wait(int sync, long timeout);
Mel Gorman0e093d992010-10-26 14:21:45 -0700310long wait_iff_congested(struct zone *zone, int sync, long timeout);
Wanpeng Li3965c9a2012-07-31 16:41:52 -0700311int pdflush_proc_obsolete(struct ctl_table *table, int write,
312 void __user *buffer, size_t *lenp, loff_t *ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313
Darrick J. Wong7d311cd2013-02-21 16:42:48 -0800314static inline bool bdi_cap_stable_pages_required(struct backing_dev_info *bdi)
315{
316 return bdi->capabilities & BDI_CAP_STABLE_WRITES;
317}
318
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700319static inline bool bdi_cap_writeback_dirty(struct backing_dev_info *bdi)
320{
321 return !(bdi->capabilities & BDI_CAP_NO_WRITEBACK);
322}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700324static inline bool bdi_cap_account_dirty(struct backing_dev_info *bdi)
325{
326 return !(bdi->capabilities & BDI_CAP_NO_ACCT_DIRTY);
327}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700329static inline bool bdi_cap_account_writeback(struct backing_dev_info *bdi)
330{
331 /* Paranoia: BDI_CAP_NO_WRITEBACK implies BDI_CAP_NO_ACCT_WB */
332 return !(bdi->capabilities & (BDI_CAP_NO_ACCT_WB |
333 BDI_CAP_NO_WRITEBACK));
334}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700336static inline bool bdi_cap_swap_backed(struct backing_dev_info *bdi)
337{
338 return bdi->capabilities & BDI_CAP_SWAP_BACKED;
339}
340
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700341static inline bool mapping_cap_writeback_dirty(struct address_space *mapping)
342{
343 return bdi_cap_writeback_dirty(mapping->backing_dev_info);
344}
345
346static inline bool mapping_cap_account_dirty(struct address_space *mapping)
347{
348 return bdi_cap_account_dirty(mapping->backing_dev_info);
349}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700351static inline bool mapping_cap_swap_backed(struct address_space *mapping)
352{
353 return bdi_cap_swap_backed(mapping->backing_dev_info);
354}
355
Jens Axboe03ba3782009-09-09 09:08:54 +0200356static inline int bdi_sched_wait(void *word)
357{
358 schedule();
359 return 0;
360}
361
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362#endif /* _LINUX_BACKING_DEV_H */