blob: aff923ae8c4b963272563759b9ac52ad55778bd0 [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 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};
36
37typedef int (congested_fn)(void *, int);
38
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -070039enum bdi_stat_item {
Peter Zijlstrac9e51e42007-10-16 23:25:47 -070040 BDI_RECLAIMABLE,
Peter Zijlstra69cb51d2007-10-16 23:25:48 -070041 BDI_WRITEBACK,
Wu Fengguangc8e28ce2011-01-23 10:07:47 -060042 BDI_DIRTIED,
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
Artem Bityutskiyecd58402010-07-25 14:29:18 +030052 unsigned long last_old_flush; /* last old data flush */
Jens Axboe03ba3782009-09-09 09:08:54 +020053
Tejun Heo839a8e82013-04-01 19:08:06 -070054 struct delayed_work dwork; /* work item used for writeback */
Artem Bityutskiyecd58402010-07-25 14:29:18 +030055 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 */
Theodore Ts'o0ae45f62015-02-02 00:37:00 -050058 struct list_head b_dirty_time; /* time stamps are dirty */
Christoph Hellwigf758eea2011-04-21 18:19:44 -060059 spinlock_t list_lock; /* protects the b_* lists */
Jens Axboe03ba3782009-09-09 09:08:54 +020060};
61
Linus Torvalds1da177e2005-04-16 15:20:36 -070062struct backing_dev_info {
Jens Axboe66f3b8e2009-09-02 09:19:46 +020063 struct list_head bdi_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 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 */
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -070069
Jens Axboed9938312009-06-12 14:45:52 +020070 char *name;
71
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -070072 struct percpu_counter bdi_stat[NR_BDI_STAT_ITEMS];
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -070073
Wu Fengguange98be2d2010-08-29 11:22:30 -060074 unsigned long bw_time_stamp; /* last time write bw is updated */
Wu Fengguangbe3ffa22011-06-12 10:51:31 -060075 unsigned long dirtied_stamp;
Wu Fengguange98be2d2010-08-29 11:22:30 -060076 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
Wu Fengguangbe3ffa22011-06-12 10:51:31 -060080 /*
81 * The base dirty throttle rate, re-calculated on every 200ms.
82 * All the bdi tasks' dirty rate will be curbed under it.
Wu Fengguang73811312011-08-26 15:53:24 -060083 * @dirty_ratelimit tracks the estimated @balanced_dirty_ratelimit
84 * in small steps and is much more smooth/stable than the latter.
Wu Fengguangbe3ffa22011-06-12 10:51:31 -060085 */
86 unsigned long dirty_ratelimit;
Wu Fengguang73811312011-08-26 15:53:24 -060087 unsigned long balanced_dirty_ratelimit;
Wu Fengguangbe3ffa22011-06-12 10:51:31 -060088
Jan Karaeb608e32012-05-24 18:59:11 +020089 struct fprop_local_percpu completions;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -070090 int dirty_exceeded;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -070091
Peter Zijlstra189d3c42008-04-30 00:54:35 -070092 unsigned int min_ratio;
Peter Zijlstraa42dde02008-04-30 00:54:36 -070093 unsigned int max_ratio, max_prop_frac;
Peter Zijlstra189d3c42008-04-30 00:54:35 -070094
Jens Axboe03ba3782009-09-09 09:08:54 +020095 struct bdi_writeback wb; /* default writeback info for this bdi */
Jan Kara5acda9d2014-04-03 14:46:23 -070096 spinlock_t wb_lock; /* protects work_list & wb.dwork scheduling */
Miklos Szeredi76f14182008-04-30 00:54:36 -070097
Jens Axboe03ba3782009-09-09 09:08:54 +020098 struct list_head work_list;
99
100 struct device *dev;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200101
Matthew Garrett31373d02010-04-06 14:25:14 +0200102 struct timer_list laptop_mode_wb_timer;
103
Miklos Szeredi76f14182008-04-30 00:54:36 -0700104#ifdef CONFIG_DEBUG_FS
105 struct dentry *debug_dir;
106 struct dentry *debug_stats;
107#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108};
109
Christoph Hellwigde1414a2015-01-14 10:42:36 +0100110struct backing_dev_info *inode_to_bdi(struct inode *inode);
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);
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100120int __must_check bdi_setup_and_register(struct backing_dev_info *, char *);
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);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700127
Jens Axboe03ba3782009-09-09 09:08:54 +0200128extern spinlock_t bdi_lock;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200129extern struct list_head bdi_list;
130
Tejun Heo839a8e82013-04-01 19:08:06 -0700131extern struct workqueue_struct *bdi_wq;
132
Jens Axboe03ba3782009-09-09 09:08:54 +0200133static 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
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700140static inline void __add_bdi_stat(struct backing_dev_info *bdi,
141 enum bdi_stat_item item, s64 amount)
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700142{
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700143 __percpu_counter_add(&bdi->bdi_stat[item], amount, BDI_STAT_BATCH);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700144}
145
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700146static inline void __inc_bdi_stat(struct backing_dev_info *bdi,
147 enum bdi_stat_item item)
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700148{
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700149 __add_bdi_stat(bdi, item, 1);
150}
151
152static 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
162static 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
168static 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
178static 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
184static 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
190static inline s64 bdi_stat_sum(struct backing_dev_info *bdi,
191 enum bdi_stat_item item)
192{
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;
201}
202
Miklos Szeredidd5656e2008-04-30 00:54:37 -0700203extern void bdi_writeout_inc(struct backing_dev_info *bdi);
204
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700205/*
206 * maximal error of a stat counter.
207 */
208static inline unsigned long bdi_stat_error(struct backing_dev_info *bdi)
209{
210#ifdef CONFIG_SMP
211 return nr_cpu_ids * BDI_STAT_BATCH;
212#else
213 return 1;
214#endif
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700215}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700217int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700218int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700219
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220/*
221 * Flags in backing_dev_info::capability
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700222 *
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
Maxim Patlasov5a537482013-09-11 14:22:46 -0700234 * BDI_CAP_STRICTLIMIT: Keep number of dirty pages below bdi threshold.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 */
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700236#define BDI_CAP_NO_ACCT_DIRTY 0x00000001
237#define BDI_CAP_NO_WRITEBACK 0x00000002
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100238#define BDI_CAP_NO_ACCT_WB 0x00000004
239#define BDI_CAP_STABLE_WRITES 0x00000008
240#define BDI_CAP_STRICTLIMIT 0x00000010
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700242#define BDI_CAP_NO_ACCT_AND_WRITEBACK \
243 (BDI_CAP_NO_WRITEBACK | BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_ACCT_WB)
244
Jörn Engel5129a462010-04-25 08:54:42 +0200245extern struct backing_dev_info noop_backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247int writeback_in_progress(struct backing_dev_info *bdi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248
249static 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
256static inline int bdi_read_congested(struct backing_dev_info *bdi)
257{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200258 return bdi_congested(bdi, 1 << BDI_sync_congested);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259}
260
261static inline int bdi_write_congested(struct backing_dev_info *bdi)
262{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200263 return bdi_congested(bdi, 1 << BDI_async_congested);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264}
265
266static inline int bdi_rw_congested(struct backing_dev_info *bdi)
267{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200268 return bdi_congested(bdi, (1 << BDI_sync_congested) |
269 (1 << BDI_async_congested));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270}
271
Trond Myklebust373c0a72009-07-11 10:06:54 -0400272enum {
273 BLK_RW_ASYNC = 0,
274 BLK_RW_SYNC = 1,
275};
276
Jens Axboe8aa7e842009-07-09 14:52:32 +0200277void clear_bdi_congested(struct backing_dev_info *bdi, int sync);
278void set_bdi_congested(struct backing_dev_info *bdi, int sync);
279long congestion_wait(int sync, long timeout);
Mel Gorman0e093d992010-10-26 14:21:45 -0700280long wait_iff_congested(struct zone *zone, int sync, long timeout);
Wanpeng Li3965c9a2012-07-31 16:41:52 -0700281int pdflush_proc_obsolete(struct ctl_table *table, int write,
282 void __user *buffer, size_t *lenp, loff_t *ppos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283
Darrick J. Wong7d311cd2013-02-21 16:42:48 -0800284static inline bool bdi_cap_stable_pages_required(struct backing_dev_info *bdi)
285{
286 return bdi->capabilities & BDI_CAP_STABLE_WRITES;
287}
288
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700289static inline bool bdi_cap_writeback_dirty(struct backing_dev_info *bdi)
290{
291 return !(bdi->capabilities & BDI_CAP_NO_WRITEBACK);
292}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700294static inline bool bdi_cap_account_dirty(struct backing_dev_info *bdi)
295{
296 return !(bdi->capabilities & BDI_CAP_NO_ACCT_DIRTY);
297}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700299static 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}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700306static inline bool mapping_cap_writeback_dirty(struct address_space *mapping)
307{
Christoph Hellwigde1414a2015-01-14 10:42:36 +0100308 return bdi_cap_writeback_dirty(inode_to_bdi(mapping->host));
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700309}
310
311static inline bool mapping_cap_account_dirty(struct address_space *mapping)
312{
Christoph Hellwigde1414a2015-01-14 10:42:36 +0100313 return bdi_cap_account_dirty(inode_to_bdi(mapping->host));
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700314}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315
Jens Axboe03ba3782009-09-09 09:08:54 +0200316static inline int bdi_sched_wait(void *word)
317{
318 schedule();
319 return 0;
320}
321
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322#endif /* _LINUX_BACKING_DEV_H */