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/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
  21 struct page;
  22 struct device;
  23 struct dentry;
  24
  25 /*
  26  * Bits in backing_dev_info.state
  27  */
  28 enum bdi_state {
  29         BDI_pending,            /* On its way to being activated */
  30         BDI_wb_alloc,           /* Default embedded wb allocated */
  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         BDI_unused,             /* Available bits start here */
  36 };
  37
  38 typedef int (congested_fn)(void *, int);
  39
  40 enum bdi_stat_item {
  41         BDI_RECLAIMABLE,
  42         BDI_WRITEBACK,
  43         BDI_DIRTIED,
  44         BDI_WRITTEN,
  45         NR_BDI_STAT_ITEMS
  46 };
  47
  48 #define BDI_STAT_BATCH (8*(1+ilog2(nr_cpu_ids)))
  49
  50 struct bdi_writeback {
  51         struct backing_dev_info *bdi;   /* our parent bdi */
  52         unsigned int nr;
  53
  54         unsigned long last_old_flush;   /* last old data flush */
  55         unsigned long last_active;      /* last time bdi thread was active */
  56
  57         struct task_struct *task;       /* writeback thread */
  58         struct timer_list wakeup_timer; /* used for delayed bdi thread wakeup */
  59         struct list_head b_dirty;       /* dirty inodes */
  60         struct list_head b_io;          /* parked for writeback */
  61         struct list_head b_more_io;     /* parked for more writeback */
  62         spinlock_t list_lock;           /* protects the b_* lists */
  63 };
  64
  65 struct backing_dev_info {
  66         struct list_head bdi_list;
  67         unsigned long ra_pages; /* max readahead in PAGE_CACHE_SIZE units */
  68         unsigned long state;    /* Always use atomic bitops on this */
  69         unsigned int capabilities; /* Device capabilities */
  70         congested_fn *congested_fn; /* Function pointer if device is md/dm */
  71         void *congested_data;   /* Pointer to aux data for congested func */
  72
  73         char *name;
  74
  75         struct percpu_counter bdi_stat[NR_BDI_STAT_ITEMS];
  76
  77         unsigned long bw_time_stamp;    /* last time write bw is updated */
  78         unsigned long dirtied_stamp;
  79         unsigned long written_stamp;    /* pages written at bw_time_stamp */
  80         unsigned long write_bandwidth;  /* the estimated write bandwidth */
  81         unsigned long avg_write_bandwidth; /* further smoothed write bw */
  82
  83         /*
  84          * The base dirty throttle rate, re-calculated on every 200ms.
  85          * All the bdi tasks' dirty rate will be curbed under it.
  86          * @dirty_ratelimit tracks the estimated @balanced_dirty_ratelimit
  87          * in small steps and is much more smooth/stable than the latter.
  88          */
  89         unsigned long dirty_ratelimit;
  90         unsigned long balanced_dirty_ratelimit;
  91
  92         struct prop_local_percpu completions;
  93         int dirty_exceeded;
  94
  95         unsigned int min_ratio;
  96         unsigned int max_ratio, max_prop_frac;
  97
  98         struct bdi_writeback wb;  /* default writeback info for this bdi */
  99         spinlock_t wb_lock;       /* protects work_list */
 100
 101         struct list_head work_list;
 102
 103         struct device *dev;
 104
 105         struct timer_list laptop_mode_wb_timer;
 106
 107 #ifdef CONFIG_DEBUG_FS
 108         struct dentry *debug_dir;
 109         struct dentry *debug_stats;
 110 #endif
 111 };
 112
 113 int bdi_init(struct backing_dev_info *bdi);
 114 void bdi_destroy(struct backing_dev_info *bdi);
 115
 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);
 120 int bdi_setup_and_register(struct backing_dev_info *, char *, unsigned int);
 121 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages);
 122 void bdi_start_background_writeback(struct backing_dev_info *bdi);
 123 int bdi_writeback_thread(void *data);
 124 int bdi_has_dirty_io(struct backing_dev_info *bdi);
 125 void bdi_arm_supers_timer(void);
 126 void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi);
 127 void bdi_lock_two(struct bdi_writeback *wb1, struct bdi_writeback *wb2);
 128
 129 extern spinlock_t bdi_lock;
 130 extern struct list_head bdi_list;
 131 extern struct list_head bdi_pending_list;
 132
 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
 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)
 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
 203 extern void bdi_writeout_inc(struct backing_dev_info *bdi);
 204
 205 /*
 206  * maximal error of a stat counter.
 207  */
 208 static 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
 215 }
 216
 217 int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
 218 int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
 219
 220 /*
 221  * Flags in backing_dev_info::capability
 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
 234  *
 235  * These flags let !MMU mmap() govern direct device mapping vs immediate
 236  * copying more easily for MAP_PRIVATE, especially for ROM filesystems.
 237  *
 238  * BDI_CAP_MAP_COPY:       Copy can be mapped (MAP_PRIVATE)
 239  * BDI_CAP_MAP_DIRECT:     Can be mapped directly (MAP_SHARED)
 240  * BDI_CAP_READ_MAP:       Can be mapped for reading
 241  * BDI_CAP_WRITE_MAP:      Can be mapped for writing
 242  * BDI_CAP_EXEC_MAP:       Can be mapped for execution
 243  *
 244  * BDI_CAP_SWAP_BACKED:    Count shmem/tmpfs objects as swap-backed.
 245  */
 246 #define BDI_CAP_NO_ACCT_DIRTY   0x00000001
 247 #define BDI_CAP_NO_WRITEBACK    0x00000002
 248 #define BDI_CAP_MAP_COPY        0x00000004
 249 #define BDI_CAP_MAP_DIRECT      0x00000008
 250 #define BDI_CAP_READ_MAP        0x00000010
 251 #define BDI_CAP_WRITE_MAP       0x00000020
 252 #define BDI_CAP_EXEC_MAP        0x00000040
 253 #define BDI_CAP_NO_ACCT_WB      0x00000080
 254 #define BDI_CAP_SWAP_BACKED     0x00000100
 255
 256 #define BDI_CAP_VMFLAGS \
 257         (BDI_CAP_READ_MAP | BDI_CAP_WRITE_MAP | BDI_CAP_EXEC_MAP)
 258
 259 #define BDI_CAP_NO_ACCT_AND_WRITEBACK \
 260         (BDI_CAP_NO_WRITEBACK | BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_ACCT_WB)
 261
 262 #if defined(VM_MAYREAD) && \
 263         (BDI_CAP_READ_MAP != VM_MAYREAD || \
 264          BDI_CAP_WRITE_MAP != VM_MAYWRITE || \
 265          BDI_CAP_EXEC_MAP != VM_MAYEXEC)
 266 #error please change backing_dev_info::capabilities flags
 267 #endif
 268
 269 extern struct backing_dev_info default_backing_dev_info;
 270 extern struct backing_dev_info noop_backing_dev_info;
 271
 272 int writeback_in_progress(struct backing_dev_info *bdi);
 273
 274 static inline int bdi_congested(struct backing_dev_info *bdi, int bdi_bits)
 275 {
 276         if (bdi->congested_fn)
 277                 return bdi->congested_fn(bdi->congested_data, bdi_bits);
 278         return (bdi->state & bdi_bits);
 279 }
 280
 281 static inline int bdi_read_congested(struct backing_dev_info *bdi)
 282 {
 283         return bdi_congested(bdi, 1 << BDI_sync_congested);
 284 }
 285
 286 static inline int bdi_write_congested(struct backing_dev_info *bdi)
 287 {
 288         return bdi_congested(bdi, 1 << BDI_async_congested);
 289 }
 290
 291 static inline int bdi_rw_congested(struct backing_dev_info *bdi)
 292 {
 293         return bdi_congested(bdi, (1 << BDI_sync_congested) |
 294                                   (1 << BDI_async_congested));
 295 }
 296
 297 enum {
 298         BLK_RW_ASYNC    = 0,
 299         BLK_RW_SYNC     = 1,
 300 };
 301
 302 void clear_bdi_congested(struct backing_dev_info *bdi, int sync);
 303 void set_bdi_congested(struct backing_dev_info *bdi, int sync);
 304 long congestion_wait(int sync, long timeout);
 305 long wait_iff_congested(struct zone *zone, int sync, long timeout);
 306
 307 static inline bool bdi_cap_writeback_dirty(struct backing_dev_info *bdi)
 308 {
 309         return !(bdi->capabilities & BDI_CAP_NO_WRITEBACK);
 310 }
 311
 312 static inline bool bdi_cap_account_dirty(struct backing_dev_info *bdi)
 313 {
 314         return !(bdi->capabilities & BDI_CAP_NO_ACCT_DIRTY);
 315 }
 316
 317 static inline bool bdi_cap_account_writeback(struct backing_dev_info *bdi)
 318 {
 319         /* Paranoia: BDI_CAP_NO_WRITEBACK implies BDI_CAP_NO_ACCT_WB */
 320         return !(bdi->capabilities & (BDI_CAP_NO_ACCT_WB |
 321                                       BDI_CAP_NO_WRITEBACK));
 322 }
 323
 324 static inline bool bdi_cap_swap_backed(struct backing_dev_info *bdi)
 325 {
 326         return bdi->capabilities & BDI_CAP_SWAP_BACKED;
 327 }
 328
 329 static inline bool bdi_cap_flush_forker(struct backing_dev_info *bdi)
 330 {
 331         return bdi == &default_backing_dev_info;
 332 }
 333
 334 static inline bool mapping_cap_writeback_dirty(struct address_space *mapping)
 335 {
 336         return bdi_cap_writeback_dirty(mapping->backing_dev_info);
 337 }
 338
 339 static inline bool mapping_cap_account_dirty(struct address_space *mapping)
 340 {
 341         return bdi_cap_account_dirty(mapping->backing_dev_info);
 342 }
 343
 344 static inline bool mapping_cap_swap_backed(struct address_space *mapping)
 345 {
 346         return bdi_cap_swap_backed(mapping->backing_dev_info);
 347 }
 348
 349 static inline int bdi_sched_wait(void *word)
 350 {
 351         schedule();
 352         return 0;
 353 }
 354
 355 #endif          /* _LINUX_BACKING_DEV_H */