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
  22 struct page;
  23 struct device;
  24 struct dentry;
  25
  26 /*
  27  * Bits in backing_dev_info.state
  28  */
  29 enum bdi_state {
  30         BDI_pending,            /* On its way to being activated */
  31         BDI_wb_alloc,           /* Default embedded wb allocated */
  32         BDI_async_congested,    /* The async (write) queue is getting full */
  33         BDI_sync_congested,     /* The sync queue is getting full */
  34         BDI_registered,         /* bdi_register() was done */
  35         BDI_writeback_running,  /* Writeback is in progress */
  36         BDI_unused,             /* Available bits start here */
  37 };
  38
  39 typedef int (congested_fn)(void *, int);
  40
  41 enum bdi_stat_item {
  42         BDI_RECLAIMABLE,
  43         BDI_WRITEBACK,
  44         BDI_DIRTIED,
  45         BDI_WRITTEN,
  46         NR_BDI_STAT_ITEMS
  47 };
  48
  49 #define BDI_STAT_BATCH (8*(1+ilog2(nr_cpu_ids)))
  50
  51 struct bdi_writeback {
  52         struct backing_dev_info *bdi;   /* our parent bdi */
  53         unsigned int nr;
  54
  55         unsigned long last_old_flush;   /* last old data flush */
  56         unsigned long last_active;      /* last time bdi thread was active */
  57
  58         struct task_struct *task;       /* writeback thread */
  59         struct timer_list wakeup_timer; /* used for delayed bdi thread wakeup */
  60         struct list_head b_dirty;       /* dirty inodes */
  61         struct list_head b_io;          /* parked for writeback */
  62         struct list_head b_more_io;     /* parked for more writeback */
  63         spinlock_t list_lock;           /* protects the b_* lists */
  64 };
  65
  66 struct backing_dev_info {
  67         struct list_head bdi_list;
  68         unsigned long ra_pages; /* max readahead in PAGE_CACHE_SIZE units */
  69         unsigned long state;    /* Always use atomic bitops on this */
  70         unsigned int capabilities; /* Device capabilities */
  71         congested_fn *congested_fn; /* Function pointer if device is md/dm */
  72         void *congested_data;   /* Pointer to aux data for congested func */
  73
  74         char *name;
  75
  76         struct percpu_counter bdi_stat[NR_BDI_STAT_ITEMS];
  77
  78         unsigned long bw_time_stamp;    /* last time write bw is updated */
  79         unsigned long dirtied_stamp;
  80         unsigned long written_stamp;    /* pages written at bw_time_stamp */
  81         unsigned long write_bandwidth;  /* the estimated write bandwidth */
  82         unsigned long avg_write_bandwidth; /* further smoothed write bw */
  83
  84         /*
  85          * The base dirty throttle rate, re-calculated on every 200ms.
  86          * All the bdi tasks' dirty rate will be curbed under it.
  87          * @dirty_ratelimit tracks the estimated @balanced_dirty_ratelimit
  88          * in small steps and is much more smooth/stable than the latter.
  89          */
  90         unsigned long dirty_ratelimit;
  91         unsigned long balanced_dirty_ratelimit;
  92
  93         struct fprop_local_percpu completions;
  94         int dirty_exceeded;
  95
  96         unsigned int min_ratio;
  97         unsigned int max_ratio, max_prop_frac;
  98
  99         struct bdi_writeback wb;  /* default writeback info for this bdi */
 100         spinlock_t wb_lock;       /* protects work_list */
 101
 102         struct list_head work_list;
 103
 104         struct device *dev;
 105
 106         struct timer_list laptop_mode_wb_timer;
 107
 108 #ifdef CONFIG_DEBUG_FS
 109         struct dentry *debug_dir;
 110         struct dentry *debug_stats;
 111 #endif
 112 };
 113
 114 int bdi_init(struct backing_dev_info *bdi);
 115 void bdi_destroy(struct backing_dev_info *bdi);
 116
 117 __printf(3, 4)
 118 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
 119                 const char *fmt, ...);
 120 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev);
 121 void bdi_unregister(struct backing_dev_info *bdi);
 122 int bdi_setup_and_register(struct backing_dev_info *, char *, unsigned int);
 123 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
 124                         enum wb_reason reason);
 125 void bdi_start_background_writeback(struct backing_dev_info *bdi);
 126 int bdi_writeback_thread(void *data);
 127 int bdi_has_dirty_io(struct backing_dev_info *bdi);
 128 void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi);
 129 void bdi_lock_two(struct bdi_writeback *wb1, struct bdi_writeback *wb2);
 130
 131 extern spinlock_t bdi_lock;
 132 extern struct list_head bdi_list;
 133 extern struct list_head bdi_pending_list;
 134
 135 static inline int wb_has_dirty_io(struct bdi_writeback *wb)
 136 {
 137         return !list_empty(&wb->b_dirty) ||
 138                !list_empty(&wb->b_io) ||
 139                !list_empty(&wb->b_more_io);
 140 }
 141
 142 static inline void __add_bdi_stat(struct backing_dev_info *bdi,
 143                 enum bdi_stat_item item, s64 amount)
 144 {
 145         __percpu_counter_add(&bdi->bdi_stat[item], amount, BDI_STAT_BATCH);
 146 }
 147
 148 static inline void __inc_bdi_stat(struct backing_dev_info *bdi,
 149                 enum bdi_stat_item item)
 150 {
 151         __add_bdi_stat(bdi, item, 1);
 152 }
 153
 154 static inline void inc_bdi_stat(struct backing_dev_info *bdi,
 155                 enum bdi_stat_item item)
 156 {
 157         unsigned long flags;
 158
 159         local_irq_save(flags);
 160         __inc_bdi_stat(bdi, item);
 161         local_irq_restore(flags);
 162 }
 163
 164 static inline void __dec_bdi_stat(struct backing_dev_info *bdi,
 165                 enum bdi_stat_item item)
 166 {
 167         __add_bdi_stat(bdi, item, -1);
 168 }
 169
 170 static inline void dec_bdi_stat(struct backing_dev_info *bdi,
 171                 enum bdi_stat_item item)
 172 {
 173         unsigned long flags;
 174
 175         local_irq_save(flags);
 176         __dec_bdi_stat(bdi, item);
 177         local_irq_restore(flags);
 178 }
 179
 180 static inline s64 bdi_stat(struct backing_dev_info *bdi,
 181                 enum bdi_stat_item item)
 182 {
 183         return percpu_counter_read_positive(&bdi->bdi_stat[item]);
 184 }
 185
 186 static inline s64 __bdi_stat_sum(struct backing_dev_info *bdi,
 187                 enum bdi_stat_item item)
 188 {
 189         return percpu_counter_sum_positive(&bdi->bdi_stat[item]);
 190 }
 191
 192 static inline s64 bdi_stat_sum(struct backing_dev_info *bdi,
 193                 enum bdi_stat_item item)
 194 {
 195         s64 sum;
 196         unsigned long flags;
 197
 198         local_irq_save(flags);
 199         sum = __bdi_stat_sum(bdi, item);
 200         local_irq_restore(flags);
 201
 202         return sum;
 203 }
 204
 205 extern void bdi_writeout_inc(struct backing_dev_info *bdi);
 206
 207 /*
 208  * maximal error of a stat counter.
 209  */
 210 static inline unsigned long bdi_stat_error(struct backing_dev_info *bdi)
 211 {
 212 #ifdef CONFIG_SMP
 213         return nr_cpu_ids * BDI_STAT_BATCH;
 214 #else
 215         return 1;
 216 #endif
 217 }
 218
 219 int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
 220 int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
 221
 222 /*
 223  * Flags in backing_dev_info::capability
 224  *
 225  * The first three flags control whether dirty pages will contribute to the
 226  * VM's accounting and whether writepages() should be called for dirty pages
 227  * (something that would not, for example, be appropriate for ramfs)
 228  *
 229  * WARNING: these flags are closely related and should not normally be
 230  * used separately.  The BDI_CAP_NO_ACCT_AND_WRITEBACK combines these
 231  * three flags into a single convenience macro.
 232  *
 233  * BDI_CAP_NO_ACCT_DIRTY:  Dirty pages shouldn't contribute to accounting
 234  * BDI_CAP_NO_WRITEBACK:   Don't write pages back
 235  * BDI_CAP_NO_ACCT_WB:     Don't automatically account writeback pages
 236  *
 237  * These flags let !MMU mmap() govern direct device mapping vs immediate
 238  * copying more easily for MAP_PRIVATE, especially for ROM filesystems.
 239  *
 240  * BDI_CAP_MAP_COPY:       Copy can be mapped (MAP_PRIVATE)
 241  * BDI_CAP_MAP_DIRECT:     Can be mapped directly (MAP_SHARED)
 242  * BDI_CAP_READ_MAP:       Can be mapped for reading
 243  * BDI_CAP_WRITE_MAP:      Can be mapped for writing
 244  * BDI_CAP_EXEC_MAP:       Can be mapped for execution
 245  *
 246  * BDI_CAP_SWAP_BACKED:    Count shmem/tmpfs objects as swap-backed.
 247  */
 248 #define BDI_CAP_NO_ACCT_DIRTY   0x00000001
 249 #define BDI_CAP_NO_WRITEBACK    0x00000002
 250 #define BDI_CAP_MAP_COPY        0x00000004
 251 #define BDI_CAP_MAP_DIRECT      0x00000008
 252 #define BDI_CAP_READ_MAP        0x00000010
 253 #define BDI_CAP_WRITE_MAP       0x00000020
 254 #define BDI_CAP_EXEC_MAP        0x00000040
 255 #define BDI_CAP_NO_ACCT_WB      0x00000080
 256 #define BDI_CAP_SWAP_BACKED     0x00000100
 257 #define BDI_CAP_STABLE_WRITES   0x00000200
 258
 259 #define BDI_CAP_VMFLAGS \
 260         (BDI_CAP_READ_MAP | BDI_CAP_WRITE_MAP | BDI_CAP_EXEC_MAP)
 261
 262 #define BDI_CAP_NO_ACCT_AND_WRITEBACK \
 263         (BDI_CAP_NO_WRITEBACK | BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_ACCT_WB)
 264
 265 #if defined(VM_MAYREAD) && \
 266         (BDI_CAP_READ_MAP != VM_MAYREAD || \
 267          BDI_CAP_WRITE_MAP != VM_MAYWRITE || \
 268          BDI_CAP_EXEC_MAP != VM_MAYEXEC)
 269 #error please change backing_dev_info::capabilities flags
 270 #endif
 271
 272 extern struct backing_dev_info default_backing_dev_info;
 273 extern struct backing_dev_info noop_backing_dev_info;
 274
 275 int writeback_in_progress(struct backing_dev_info *bdi);
 276
 277 static inline int bdi_congested(struct backing_dev_info *bdi, int bdi_bits)
 278 {
 279         if (bdi->congested_fn)
 280                 return bdi->congested_fn(bdi->congested_data, bdi_bits);
 281         return (bdi->state & bdi_bits);
 282 }
 283
 284 static inline int bdi_read_congested(struct backing_dev_info *bdi)
 285 {
 286         return bdi_congested(bdi, 1 << BDI_sync_congested);
 287 }
 288
 289 static inline int bdi_write_congested(struct backing_dev_info *bdi)
 290 {
 291         return bdi_congested(bdi, 1 << BDI_async_congested);
 292 }
 293
 294 static inline int bdi_rw_congested(struct backing_dev_info *bdi)
 295 {
 296         return bdi_congested(bdi, (1 << BDI_sync_congested) |
 297                                   (1 << BDI_async_congested));
 298 }
 299
 300 enum {
 301         BLK_RW_ASYNC    = 0,
 302         BLK_RW_SYNC     = 1,
 303 };
 304
 305 void clear_bdi_congested(struct backing_dev_info *bdi, int sync);
 306 void set_bdi_congested(struct backing_dev_info *bdi, int sync);
 307 long congestion_wait(int sync, long timeout);
 308 long wait_iff_congested(struct zone *zone, int sync, long timeout);
 309 int pdflush_proc_obsolete(struct ctl_table *table, int write,
 310                 void __user *buffer, size_t *lenp, loff_t *ppos);
 311
 312 static inline bool bdi_cap_stable_pages_required(struct backing_dev_info *bdi)
 313 {
 314         return bdi->capabilities & BDI_CAP_STABLE_WRITES;
 315 }
 316
 317 static inline bool bdi_cap_writeback_dirty(struct backing_dev_info *bdi)
 318 {
 319         return !(bdi->capabilities & BDI_CAP_NO_WRITEBACK);
 320 }
 321
 322 static inline bool bdi_cap_account_dirty(struct backing_dev_info *bdi)
 323 {
 324         return !(bdi->capabilities & BDI_CAP_NO_ACCT_DIRTY);
 325 }
 326
 327 static inline bool bdi_cap_account_writeback(struct backing_dev_info *bdi)
 328 {
 329         /* Paranoia: BDI_CAP_NO_WRITEBACK implies BDI_CAP_NO_ACCT_WB */
 330         return !(bdi->capabilities & (BDI_CAP_NO_ACCT_WB |
 331                                       BDI_CAP_NO_WRITEBACK));
 332 }
 333
 334 static inline bool bdi_cap_swap_backed(struct backing_dev_info *bdi)
 335 {
 336         return bdi->capabilities & BDI_CAP_SWAP_BACKED;
 337 }
 338
 339 static inline bool bdi_cap_flush_forker(struct backing_dev_info *bdi)
 340 {
 341         return bdi == &default_backing_dev_info;
 342 }
 343
 344 static inline bool mapping_cap_writeback_dirty(struct address_space *mapping)
 345 {
 346         return bdi_cap_writeback_dirty(mapping->backing_dev_info);
 347 }
 348
 349 static inline bool mapping_cap_account_dirty(struct address_space *mapping)
 350 {
 351         return bdi_cap_account_dirty(mapping->backing_dev_info);
 352 }
 353
 354 static inline bool mapping_cap_swap_backed(struct address_space *mapping)
 355 {
 356         return bdi_cap_swap_backed(mapping->backing_dev_info);
 357 }
 358
 359 static inline int bdi_sched_wait(void *word)
 360 {
 361         schedule();
 362         return 0;
 363 }
 364
 365 #endif          /* _LINUX_BACKING_DEV_H */