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/mutex.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_pending,            /* On its way to being activated */
  32         BDI_wb_alloc,           /* Default embedded wb allocated */
  33         BDI_async_congested,    /* The async (write) queue is getting full */
  34         BDI_sync_congested,     /* The sync queue is getting full */
  35         BDI_registered,         /* bdi_register() was done */
  36         BDI_writeback_running,  /* Writeback is in progress */
  37         BDI_unused,             /* Available bits start here */
  38 };
  39
  40 typedef int (congested_fn)(void *, int);
  41
  42 enum bdi_stat_item {
  43         BDI_RECLAIMABLE,
  44         BDI_WRITEBACK,
  45         BDI_DIRTIED,
  46         BDI_WRITTEN,
  47         NR_BDI_STAT_ITEMS
  48 };
  49
  50 #define BDI_STAT_BATCH (8*(1+ilog2(nr_cpu_ids)))
  51
  52 struct bdi_writeback {
  53         struct backing_dev_info *bdi;   /* our parent bdi */
  54         unsigned int nr;
  55
  56         unsigned long last_old_flush;   /* last old data flush */
  57         unsigned long last_active;      /* last time bdi thread was active */
  58
  59         struct task_struct *task;       /* writeback thread */
  60         struct timer_list wakeup_timer; /* used for delayed bdi thread wakeup */
  61         struct list_head b_dirty;       /* dirty inodes */
  62         struct list_head b_io;          /* parked for writeback */
  63         struct list_head b_more_io;     /* parked for more writeback */
  64         spinlock_t list_lock;           /* protects the b_* lists */
  65 };
  66
  67 struct backing_dev_info {
  68         struct list_head bdi_list;
  69         unsigned long ra_pages; /* max readahead in PAGE_CACHE_SIZE units */
  70         unsigned long state;    /* Always use atomic bitops on this */
  71         unsigned int capabilities; /* Device capabilities */
  72         congested_fn *congested_fn; /* Function pointer if device is md/dm */
  73         void *congested_data;   /* Pointer to aux data for congested func */
  74
  75         char *name;
  76
  77         struct percpu_counter bdi_stat[NR_BDI_STAT_ITEMS];
  78
  79         unsigned long bw_time_stamp;    /* last time write bw is updated */
  80         unsigned long dirtied_stamp;
  81         unsigned long written_stamp;    /* pages written at bw_time_stamp */
  82         unsigned long write_bandwidth;  /* the estimated write bandwidth */
  83         unsigned long avg_write_bandwidth; /* further smoothed write bw */
  84
  85         /*
  86          * The base dirty throttle rate, re-calculated on every 200ms.
  87          * All the bdi tasks' dirty rate will be curbed under it.
  88          * @dirty_ratelimit tracks the estimated @balanced_dirty_ratelimit
  89          * in small steps and is much more smooth/stable than the latter.
  90          */
  91         unsigned long dirty_ratelimit;
  92         unsigned long balanced_dirty_ratelimit;
  93
  94         struct fprop_local_percpu completions;
  95         int dirty_exceeded;
  96
  97         unsigned int min_ratio;
  98         unsigned int max_ratio, max_prop_frac;
  99
 100         struct bdi_writeback wb;  /* default writeback info for this bdi */
 101         spinlock_t wb_lock;       /* protects work_list */
 102
 103         struct list_head work_list;
 104
 105         struct device *dev;
 106
 107         struct timer_list laptop_mode_wb_timer;
 108
 109         cpumask_t *flusher_cpumask; /* used for writeback thread scheduling */
 110         struct mutex flusher_cpumask_lock;
 111
 112 #ifdef CONFIG_DEBUG_FS
 113         struct dentry *debug_dir;
 114         struct dentry *debug_stats;
 115 #endif
 116 };
 117
 118 int bdi_init(struct backing_dev_info *bdi);
 119 void bdi_destroy(struct backing_dev_info *bdi);
 120
 121 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
 122                 const char *fmt, ...);
 123 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev);
 124 void bdi_unregister(struct backing_dev_info *bdi);
 125 int bdi_setup_and_register(struct backing_dev_info *, char *, unsigned int);
 126 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
 127                         enum wb_reason reason);
 128 void bdi_start_background_writeback(struct backing_dev_info *bdi);
 129 int bdi_writeback_thread(void *data);
 130 int bdi_has_dirty_io(struct backing_dev_info *bdi);
 131 void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi);
 132 void bdi_lock_two(struct bdi_writeback *wb1, struct bdi_writeback *wb2);
 133
 134 extern spinlock_t bdi_lock;
 135 extern struct list_head bdi_list;
 136 extern struct list_head bdi_pending_list;
 137
 138 static inline int wb_has_dirty_io(struct bdi_writeback *wb)
 139 {
 140         return !list_empty(&wb->b_dirty) ||
 141                !list_empty(&wb->b_io) ||
 142                !list_empty(&wb->b_more_io);
 143 }
 144
 145 static inline void __add_bdi_stat(struct backing_dev_info *bdi,
 146                 enum bdi_stat_item item, s64 amount)
 147 {
 148         __percpu_counter_add(&bdi->bdi_stat[item], amount, BDI_STAT_BATCH);
 149 }
 150
 151 static inline void __inc_bdi_stat(struct backing_dev_info *bdi,
 152                 enum bdi_stat_item item)
 153 {
 154         __add_bdi_stat(bdi, item, 1);
 155 }
 156
 157 static inline void inc_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         __inc_bdi_stat(bdi, item);
 164         local_irq_restore(flags);
 165 }
 166
 167 static inline void __dec_bdi_stat(struct backing_dev_info *bdi,
 168                 enum bdi_stat_item item)
 169 {
 170         __add_bdi_stat(bdi, item, -1);
 171 }
 172
 173 static inline void dec_bdi_stat(struct backing_dev_info *bdi,
 174                 enum bdi_stat_item item)
 175 {
 176         unsigned long flags;
 177
 178         local_irq_save(flags);
 179         __dec_bdi_stat(bdi, item);
 180         local_irq_restore(flags);
 181 }
 182
 183 static inline s64 bdi_stat(struct backing_dev_info *bdi,
 184                 enum bdi_stat_item item)
 185 {
 186         return percpu_counter_read_positive(&bdi->bdi_stat[item]);
 187 }
 188
 189 static inline s64 __bdi_stat_sum(struct backing_dev_info *bdi,
 190                 enum bdi_stat_item item)
 191 {
 192         return percpu_counter_sum_positive(&bdi->bdi_stat[item]);
 193 }
 194
 195 static inline s64 bdi_stat_sum(struct backing_dev_info *bdi,
 196                 enum bdi_stat_item item)
 197 {
 198         s64 sum;
 199         unsigned long flags;
 200
 201         local_irq_save(flags);
 202         sum = __bdi_stat_sum(bdi, item);
 203         local_irq_restore(flags);
 204
 205         return sum;
 206 }
 207
 208 extern void bdi_writeout_inc(struct backing_dev_info *bdi);
 209
 210 /*
 211  * maximal error of a stat counter.
 212  */
 213 static inline unsigned long bdi_stat_error(struct backing_dev_info *bdi)
 214 {
 215 #ifdef CONFIG_SMP
 216         return nr_cpu_ids * BDI_STAT_BATCH;
 217 #else
 218         return 1;
 219 #endif
 220 }
 221
 222 int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
 223 int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
 224
 225 /*
 226  * Flags in backing_dev_info::capability
 227  *
 228  * The first three flags control whether dirty pages will contribute to the
 229  * VM's accounting and whether writepages() should be called for dirty pages
 230  * (something that would not, for example, be appropriate for ramfs)
 231  *
 232  * WARNING: these flags are closely related and should not normally be
 233  * used separately.  The BDI_CAP_NO_ACCT_AND_WRITEBACK combines these
 234  * three flags into a single convenience macro.
 235  *
 236  * BDI_CAP_NO_ACCT_DIRTY:  Dirty pages shouldn't contribute to accounting
 237  * BDI_CAP_NO_WRITEBACK:   Don't write pages back
 238  * BDI_CAP_NO_ACCT_WB:     Don't automatically account writeback pages
 239  *
 240  * These flags let !MMU mmap() govern direct device mapping vs immediate
 241  * copying more easily for MAP_PRIVATE, especially for ROM filesystems.
 242  *
 243  * BDI_CAP_MAP_COPY:       Copy can be mapped (MAP_PRIVATE)
 244  * BDI_CAP_MAP_DIRECT:     Can be mapped directly (MAP_SHARED)
 245  * BDI_CAP_READ_MAP:       Can be mapped for reading
 246  * BDI_CAP_WRITE_MAP:      Can be mapped for writing
 247  * BDI_CAP_EXEC_MAP:       Can be mapped for execution
 248  *
 249  * BDI_CAP_SWAP_BACKED:    Count shmem/tmpfs objects as swap-backed.
 250  */
 251 #define BDI_CAP_NO_ACCT_DIRTY   0x00000001
 252 #define BDI_CAP_NO_WRITEBACK    0x00000002
 253 #define BDI_CAP_MAP_COPY        0x00000004
 254 #define BDI_CAP_MAP_DIRECT      0x00000008
 255 #define BDI_CAP_READ_MAP        0x00000010
 256 #define BDI_CAP_WRITE_MAP       0x00000020
 257 #define BDI_CAP_EXEC_MAP        0x00000040
 258 #define BDI_CAP_NO_ACCT_WB      0x00000080
 259 #define BDI_CAP_SWAP_BACKED     0x00000100
 260
 261 #define BDI_CAP_VMFLAGS \
 262         (BDI_CAP_READ_MAP | BDI_CAP_WRITE_MAP | BDI_CAP_EXEC_MAP)
 263
 264 #define BDI_CAP_NO_ACCT_AND_WRITEBACK \
 265         (BDI_CAP_NO_WRITEBACK | BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_ACCT_WB)
 266
 267 #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
 274 extern struct backing_dev_info default_backing_dev_info;
 275 extern struct backing_dev_info noop_backing_dev_info;
 276
 277 int writeback_in_progress(struct backing_dev_info *bdi);
 278
 279 static 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
 286 static inline int bdi_read_congested(struct backing_dev_info *bdi)
 287 {
 288         return bdi_congested(bdi, 1 << BDI_sync_congested);
 289 }
 290
 291 static inline int bdi_write_congested(struct backing_dev_info *bdi)
 292 {
 293         return bdi_congested(bdi, 1 << BDI_async_congested);
 294 }
 295
 296 static inline int bdi_rw_congested(struct backing_dev_info *bdi)
 297 {
 298         return bdi_congested(bdi, (1 << BDI_sync_congested) |
 299                                   (1 << BDI_async_congested));
 300 }
 301
 302 enum {
 303         BLK_RW_ASYNC    = 0,
 304         BLK_RW_SYNC     = 1,
 305 };
 306
 307 void clear_bdi_congested(struct backing_dev_info *bdi, int sync);
 308 void set_bdi_congested(struct backing_dev_info *bdi, int sync);
 309 long congestion_wait(int sync, long timeout);
 310 long wait_iff_congested(struct zone *zone, int sync, long timeout);
 311 int pdflush_proc_obsolete(struct ctl_table *table, int write,
 312                 void __user *buffer, size_t *lenp, loff_t *ppos);
 313
 314 static inline bool bdi_cap_writeback_dirty(struct backing_dev_info *bdi)
 315 {
 316         return !(bdi->capabilities & BDI_CAP_NO_WRITEBACK);
 317 }
 318
 319 static inline bool bdi_cap_account_dirty(struct backing_dev_info *bdi)
 320 {
 321         return !(bdi->capabilities & BDI_CAP_NO_ACCT_DIRTY);
 322 }
 323
 324 static inline bool bdi_cap_account_writeback(struct backing_dev_info *bdi)
 325 {
 326         /* Paranoia: BDI_CAP_NO_WRITEBACK implies BDI_CAP_NO_ACCT_WB */
 327         return !(bdi->capabilities & (BDI_CAP_NO_ACCT_WB |
 328                                       BDI_CAP_NO_WRITEBACK));
 329 }
 330
 331 static inline bool bdi_cap_swap_backed(struct backing_dev_info *bdi)
 332 {
 333         return bdi->capabilities & BDI_CAP_SWAP_BACKED;
 334 }
 335
 336 static inline bool bdi_cap_flush_forker(struct backing_dev_info *bdi)
 337 {
 338         return bdi == &default_backing_dev_info;
 339 }
 340
 341 static inline bool mapping_cap_writeback_dirty(struct address_space *mapping)
 342 {
 343         return bdi_cap_writeback_dirty(mapping->backing_dev_info);
 344 }
 345
 346 static inline bool mapping_cap_account_dirty(struct address_space *mapping)
 347 {
 348         return bdi_cap_account_dirty(mapping->backing_dev_info);
 349 }
 350
 351 static inline bool mapping_cap_swap_backed(struct address_space *mapping)
 352 {
 353         return bdi_cap_swap_backed(mapping->backing_dev_info);
 354 }
 355
 356 static inline int bdi_sched_wait(void *word)
 357 {
 358         schedule();
 359         return 0;
 360 }
 361
 362 #endif          /* _LINUX_BACKING_DEV_H */