include/linux/backing-dev.h

   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
  11 #include <linux/percpu_counter.h>
  12 #include <linux/log2.h>
  13 #include <linux/flex_proportions.h>
  14 #include <linux/kernel.h>
  15 #include <linux/fs.h>
  16 #include <linux/sched.h>
  17 #include <linux/timer.h>
  18 #include <linux/writeback.h>
  19 #include <linux/atomic.h>
  20 #include <linux/sysctl.h>
  21 #include <linux/workqueue.h>
  22
  23 struct page;
  24 struct device;
  25 struct dentry;
  26
  27 /*
  28  * Bits in backing_dev_info.state
  29  */
  30 enum bdi_state {
  31         BDI_async_congested,    /* The async (write) queue is getting full */
  32         BDI_sync_congested,     /* The sync queue is getting full */
  33         BDI_registered,         /* bdi_register() was done */
  34         BDI_writeback_running,  /* Writeback is in progress */
  35 };
  36
  37 typedef int (congested_fn)(void *, int);
  38
  39 enum bdi_stat_item {
  40         BDI_RECLAIMABLE,
  41         BDI_WRITEBACK,
  42         BDI_DIRTIED,
  43         BDI_WRITTEN,
  44         NR_BDI_STAT_ITEMS
  45 };
  46
  47 #define BDI_STAT_BATCH (8*(1+ilog2(nr_cpu_ids)))
  48
  49 struct bdi_writeback {
  50         struct backing_dev_info *bdi;   /* our parent bdi */
  51
  52         unsigned long last_old_flush;   /* last old data flush */
  53
  54         struct delayed_work dwork;      /* work item used for writeback */
  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 */
  58         spinlock_t list_lock;           /* protects the b_* lists */
  59 };
  60
  61 struct backing_dev_info {
  62         struct list_head bdi_list;
  63         unsigned long ra_pages; /* max readahead in PAGE_CACHE_SIZE units */
  64         unsigned long state;    /* Always use atomic bitops on this */
  65         unsigned int capabilities; /* Device capabilities */
  66         congested_fn *congested_fn; /* Function pointer if device is md/dm */
  67         void *congested_data;   /* Pointer to aux data for congested func */
  68
  69         char *name;
  70
  71         struct percpu_counter bdi_stat[NR_BDI_STAT_ITEMS];
  72
  73         unsigned long bw_time_stamp;    /* last time write bw is updated */
  74         unsigned long dirtied_stamp;
  75         unsigned long written_stamp;    /* pages written at bw_time_stamp */
  76         unsigned long write_bandwidth;  /* the estimated write bandwidth */
  77         unsigned long avg_write_bandwidth; /* further smoothed write bw */
  78
  79         /*
  80          * The base dirty throttle rate, re-calculated on every 200ms.
  81          * All the bdi tasks' dirty rate will be curbed under it.
  82          * @dirty_ratelimit tracks the estimated @balanced_dirty_ratelimit
  83          * in small steps and is much more smooth/stable than the latter.
  84          */
  85         unsigned long dirty_ratelimit;
  86         unsigned long balanced_dirty_ratelimit;
  87
  88         struct fprop_local_percpu completions;
  89         int dirty_exceeded;
  90
  91         unsigned int min_ratio;
  92         unsigned int max_ratio, max_prop_frac;
  93
  94         struct bdi_writeback wb;  /* default writeback info for this bdi */
  95         spinlock_t wb_lock;       /* protects work_list & wb.dwork scheduling */
  96
  97         struct list_head work_list;
  98
  99         struct device *dev;
 100
 101         struct timer_list laptop_mode_wb_timer;
 102
 103 #ifdef CONFIG_DEBUG_FS
 104         struct dentry *debug_dir;
 105         struct dentry *debug_stats;
 106 #endif
 107 };
 108
 109 int __must_check bdi_init(struct backing_dev_info *bdi);
 110 void bdi_destroy(struct backing_dev_info *bdi);
 111
 112 __printf(3, 4)
 113 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
 114                 const char *fmt, ...);
 115 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev);
 116 void bdi_unregister(struct backing_dev_info *bdi);
 117 int __must_check bdi_setup_and_register(struct backing_dev_info *, char *, unsigned int);
 118 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
 119                         enum wb_reason reason);
 120 void bdi_start_background_writeback(struct backing_dev_info *bdi);
 121 void bdi_writeback_workfn(struct work_struct *work);
 122 int bdi_has_dirty_io(struct backing_dev_info *bdi);
 123 void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi);
 124
 125 extern spinlock_t bdi_lock;
 126 extern struct list_head bdi_list;
 127
 128 extern struct workqueue_struct *bdi_wq;
 129
 130 static inline int wb_has_dirty_io(struct bdi_writeback *wb)
 131 {
 132         return !list_empty(&wb->b_dirty) ||
 133                !list_empty(&wb->b_io) ||
 134                !list_empty(&wb->b_more_io);
 135 }
 136
 137 static inline void __add_bdi_stat(struct backing_dev_info *bdi,
 138                 enum bdi_stat_item item, s64 amount)
 139 {
 140         __percpu_counter_add(&bdi->bdi_stat[item], amount, BDI_STAT_BATCH);
 141 }
 142
 143 static inline void __inc_bdi_stat(struct backing_dev_info *bdi,
 144                 enum bdi_stat_item item)
 145 {
 146         __add_bdi_stat(bdi, item, 1);
 147 }
 148
 149 static inline void inc_bdi_stat(struct backing_dev_info *bdi,
 150                 enum bdi_stat_item item)
 151 {
 152         unsigned long flags;
 153
 154         local_irq_save(flags);
 155         __inc_bdi_stat(bdi, item);
 156         local_irq_restore(flags);
 157 }
 158
 159 static inline void __dec_bdi_stat(struct backing_dev_info *bdi,
 160                 enum bdi_stat_item item)
 161 {
 162         __add_bdi_stat(bdi, item, -1);
 163 }
 164
 165 static inline void dec_bdi_stat(struct backing_dev_info *bdi,
 166                 enum bdi_stat_item item)
 167 {
 168         unsigned long flags;
 169
 170         local_irq_save(flags);
 171         __dec_bdi_stat(bdi, item);
 172         local_irq_restore(flags);
 173 }
 174
 175 static inline s64 bdi_stat(struct backing_dev_info *bdi,
 176                 enum bdi_stat_item item)
 177 {
 178         return percpu_counter_read_positive(&bdi->bdi_stat[item]);
 179 }
 180
 181 static inline s64 __bdi_stat_sum(struct backing_dev_info *bdi,
 182                 enum bdi_stat_item item)
 183 {
 184         return percpu_counter_sum_positive(&bdi->bdi_stat[item]);
 185 }
 186
 187 static inline s64 bdi_stat_sum(struct backing_dev_info *bdi,
 188                 enum bdi_stat_item item)
 189 {
 190         s64 sum;
 191         unsigned long flags;
 192
 193         local_irq_save(flags);
 194         sum = __bdi_stat_sum(bdi, item);
 195         local_irq_restore(flags);
 196
 197         return sum;
 198 }
 199
 200 extern void bdi_writeout_inc(struct backing_dev_info *bdi);
 201
 202 /*
 203  * maximal error of a stat counter.
 204  */
 205 static inline unsigned long bdi_stat_error(struct backing_dev_info *bdi)
 206 {
 207 #ifdef CONFIG_SMP
 208         return nr_cpu_ids * BDI_STAT_BATCH;
 209 #else
 210         return 1;
 211 #endif
 212 }
 213
 214 int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
 215 int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
 216
 217 /*
 218  * Flags in backing_dev_info::capability
 219  *
 220  * The first three flags control whether dirty pages will contribute to the
 221  * VM's accounting and whether writepages() should be called for dirty pages
 222  * (something that would not, for example, be appropriate for ramfs)
 223  *
 224  * WARNING: these flags are closely related and should not normally be
 225  * used separately.  The BDI_CAP_NO_ACCT_AND_WRITEBACK combines these
 226  * three flags into a single convenience macro.
 227  *
 228  * BDI_CAP_NO_ACCT_DIRTY:  Dirty pages shouldn't contribute to accounting
 229  * BDI_CAP_NO_WRITEBACK:   Don't write pages back
 230  * BDI_CAP_NO_ACCT_WB:     Don't automatically account writeback pages
 231  *
 232  * These flags let !MMU mmap() govern direct device mapping vs immediate
 233  * copying more easily for MAP_PRIVATE, especially for ROM filesystems.
 234  *
 235  * BDI_CAP_MAP_COPY:       Copy can be mapped (MAP_PRIVATE)
 236  * BDI_CAP_MAP_DIRECT:     Can be mapped directly (MAP_SHARED)
 237  * BDI_CAP_READ_MAP:       Can be mapped for reading
 238  * BDI_CAP_WRITE_MAP:      Can be mapped for writing
 239  * BDI_CAP_EXEC_MAP:       Can be mapped for execution
 240  *
 241  * BDI_CAP_STRICTLIMIT:    Keep number of dirty pages below bdi threshold.
 242  */
 243 #define BDI_CAP_NO_ACCT_DIRTY   0x00000001
 244 #define BDI_CAP_NO_WRITEBACK    0x00000002
 245 #define BDI_CAP_MAP_COPY        0x00000004
 246 #define BDI_CAP_MAP_DIRECT      0x00000008
 247 #define BDI_CAP_READ_MAP        0x00000010
 248 #define BDI_CAP_WRITE_MAP       0x00000020
 249 #define BDI_CAP_EXEC_MAP        0x00000040
 250 #define BDI_CAP_NO_ACCT_WB      0x00000080
 251 #define BDI_CAP_STABLE_WRITES   0x00000200
 252 #define BDI_CAP_STRICTLIMIT     0x00000400
 253
 254 #define BDI_CAP_VMFLAGS \
 255         (BDI_CAP_READ_MAP | BDI_CAP_WRITE_MAP | BDI_CAP_EXEC_MAP)
 256
 257 #define BDI_CAP_NO_ACCT_AND_WRITEBACK \
 258         (BDI_CAP_NO_WRITEBACK | BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_ACCT_WB)
 259
 260 #if defined(VM_MAYREAD) && \
 261         (BDI_CAP_READ_MAP != VM_MAYREAD || \
 262          BDI_CAP_WRITE_MAP != VM_MAYWRITE || \
 263          BDI_CAP_EXEC_MAP != VM_MAYEXEC)
 264 #error please change backing_dev_info::capabilities flags
 265 #endif
 266
 267 extern struct backing_dev_info default_backing_dev_info;
 268 extern struct backing_dev_info noop_backing_dev_info;
 269
 270 int writeback_in_progress(struct backing_dev_info *bdi);
 271
 272 static inline int bdi_congested(struct backing_dev_info *bdi, int bdi_bits)
 273 {
 274         if (bdi->congested_fn)
 275                 return bdi->congested_fn(bdi->congested_data, bdi_bits);
 276         return (bdi->state & bdi_bits);
 277 }
 278
 279 static inline int bdi_read_congested(struct backing_dev_info *bdi)
 280 {
 281         return bdi_congested(bdi, 1 << BDI_sync_congested);
 282 }
 283
 284 static inline int bdi_write_congested(struct backing_dev_info *bdi)
 285 {
 286         return bdi_congested(bdi, 1 << BDI_async_congested);
 287 }
 288
 289 static inline int bdi_rw_congested(struct backing_dev_info *bdi)
 290 {
 291         return bdi_congested(bdi, (1 << BDI_sync_congested) |
 292                                   (1 << BDI_async_congested));
 293 }
 294
 295 enum {
 296         BLK_RW_ASYNC    = 0,
 297         BLK_RW_SYNC     = 1,
 298 };
 299
 300 void clear_bdi_congested(struct backing_dev_info *bdi, int sync);
 301 void set_bdi_congested(struct backing_dev_info *bdi, int sync);
 302 long congestion_wait(int sync, long timeout);
 303 long wait_iff_congested(struct zone *zone, int sync, long timeout);
 304 int pdflush_proc_obsolete(struct ctl_table *table, int write,
 305                 void __user *buffer, size_t *lenp, loff_t *ppos);
 306
 307 static inline bool bdi_cap_stable_pages_required(struct backing_dev_info *bdi)
 308 {
 309         return bdi->capabilities & BDI_CAP_STABLE_WRITES;
 310 }
 311
 312 static inline bool bdi_cap_writeback_dirty(struct backing_dev_info *bdi)
 313 {
 314         return !(bdi->capabilities & BDI_CAP_NO_WRITEBACK);
 315 }
 316
 317 static inline bool bdi_cap_account_dirty(struct backing_dev_info *bdi)
 318 {
 319         return !(bdi->capabilities & BDI_CAP_NO_ACCT_DIRTY);
 320 }
 321
 322 static inline bool bdi_cap_account_writeback(struct backing_dev_info *bdi)
 323 {
 324         /* Paranoia: BDI_CAP_NO_WRITEBACK implies BDI_CAP_NO_ACCT_WB */
 325         return !(bdi->capabilities & (BDI_CAP_NO_ACCT_WB |
 326                                       BDI_CAP_NO_WRITEBACK));
 327 }
 328
 329 static inline bool mapping_cap_writeback_dirty(struct address_space *mapping)
 330 {
 331         return bdi_cap_writeback_dirty(mapping->backing_dev_info);
 332 }
 333
 334 static inline bool mapping_cap_account_dirty(struct address_space *mapping)
 335 {
 336         return bdi_cap_account_dirty(mapping->backing_dev_info);
 337 }
 338
 339 static inline int bdi_sched_wait(void *word)
 340 {
 341         schedule();
 342         return 0;
 343 }
 344
 345 #endif          /* _LINUX_BACKING_DEV_H */