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_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
  57         struct delayed_work dwork;      /* work item used for writeback */
  58         struct list_head b_dirty;       /* dirty inodes */
  59         struct list_head b_io;          /* parked for writeback */
  60         struct list_head b_more_io;     /* parked for more writeback */
  61         spinlock_t list_lock;           /* protects the b_* lists */
  62 };
  63
  64 struct backing_dev_info {
  65         struct list_head bdi_list;
  66         unsigned long ra_pages; /* max readahead in PAGE_CACHE_SIZE units */
  67         unsigned long state;    /* Always use atomic bitops on this */
  68         unsigned int capabilities; /* Device capabilities */
  69         congested_fn *congested_fn; /* Function pointer if device is md/dm */
  70         void *congested_data;   /* Pointer to aux data for congested func */
  71
  72         char *name;
  73
  74         struct percpu_counter bdi_stat[NR_BDI_STAT_ITEMS];
  75
  76         unsigned long bw_time_stamp;    /* last time write bw is updated */
  77         unsigned long dirtied_stamp;
  78         unsigned long written_stamp;    /* pages written at bw_time_stamp */
  79         unsigned long write_bandwidth;  /* the estimated write bandwidth */
  80         unsigned long avg_write_bandwidth; /* further smoothed write bw */
  81
  82         /*
  83          * The base dirty throttle rate, re-calculated on every 200ms.
  84          * All the bdi tasks' dirty rate will be curbed under it.
  85          * @dirty_ratelimit tracks the estimated @balanced_dirty_ratelimit
  86          * in small steps and is much more smooth/stable than the latter.
  87          */
  88         unsigned long dirty_ratelimit;
  89         unsigned long balanced_dirty_ratelimit;
  90
  91         struct fprop_local_percpu completions;
  92         int dirty_exceeded;
  93
  94         unsigned int min_ratio;
  95         unsigned int max_ratio, max_prop_frac;
  96
  97         struct bdi_writeback wb;  /* default writeback info for this bdi */
  98         spinlock_t wb_lock;       /* protects work_list & wb.dwork scheduling */
  99
 100         struct list_head work_list;
 101
 102         struct device *dev;
 103
 104         struct timer_list laptop_mode_wb_timer;
 105
 106 #ifdef CONFIG_DEBUG_FS
 107         struct dentry *debug_dir;
 108         struct dentry *debug_stats;
 109 #endif
 110 };
 111
 112 int __must_check bdi_init(struct backing_dev_info *bdi);
 113 void bdi_destroy(struct backing_dev_info *bdi);
 114
 115 __printf(3, 4)
 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 __must_check 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                         enum wb_reason reason);
 123 void bdi_start_background_writeback(struct backing_dev_info *bdi);
 124 void bdi_writeback_workfn(struct work_struct *work);
 125 int bdi_has_dirty_io(struct backing_dev_info *bdi);
 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
 132 extern struct workqueue_struct *bdi_wq;
 133
 134 static inline int wb_has_dirty_io(struct bdi_writeback *wb)
 135 {
 136         return !list_empty(&wb->b_dirty) ||
 137                !list_empty(&wb->b_io) ||
 138                !list_empty(&wb->b_more_io);
 139 }
 140
 141 static inline void __add_bdi_stat(struct backing_dev_info *bdi,
 142                 enum bdi_stat_item item, s64 amount)
 143 {
 144         __percpu_counter_add(&bdi->bdi_stat[item], amount, BDI_STAT_BATCH);
 145 }
 146
 147 static inline void __inc_bdi_stat(struct backing_dev_info *bdi,
 148                 enum bdi_stat_item item)
 149 {
 150         __add_bdi_stat(bdi, item, 1);
 151 }
 152
 153 static inline void inc_bdi_stat(struct backing_dev_info *bdi,
 154                 enum bdi_stat_item item)
 155 {
 156         unsigned long flags;
 157
 158         local_irq_save(flags);
 159         __inc_bdi_stat(bdi, item);
 160         local_irq_restore(flags);
 161 }
 162
 163 static inline void __dec_bdi_stat(struct backing_dev_info *bdi,
 164                 enum bdi_stat_item item)
 165 {
 166         __add_bdi_stat(bdi, item, -1);
 167 }
 168
 169 static inline void dec_bdi_stat(struct backing_dev_info *bdi,
 170                 enum bdi_stat_item item)
 171 {
 172         unsigned long flags;
 173
 174         local_irq_save(flags);
 175         __dec_bdi_stat(bdi, item);
 176         local_irq_restore(flags);
 177 }
 178
 179 static inline s64 bdi_stat(struct backing_dev_info *bdi,
 180                 enum bdi_stat_item item)
 181 {
 182         return percpu_counter_read_positive(&bdi->bdi_stat[item]);
 183 }
 184
 185 static inline s64 __bdi_stat_sum(struct backing_dev_info *bdi,
 186                 enum bdi_stat_item item)
 187 {
 188         return percpu_counter_sum_positive(&bdi->bdi_stat[item]);
 189 }
 190
 191 static inline s64 bdi_stat_sum(struct backing_dev_info *bdi,
 192                 enum bdi_stat_item item)
 193 {
 194         s64 sum;
 195         unsigned long flags;
 196
 197         local_irq_save(flags);
 198         sum = __bdi_stat_sum(bdi, item);
 199         local_irq_restore(flags);
 200
 201         return sum;
 202 }
 203
 204 extern void bdi_writeout_inc(struct backing_dev_info *bdi);
 205
 206 /*
 207  * maximal error of a stat counter.
 208  */
 209 static inline unsigned long bdi_stat_error(struct backing_dev_info *bdi)
 210 {
 211 #ifdef CONFIG_SMP
 212         return nr_cpu_ids * BDI_STAT_BATCH;
 213 #else
 214         return 1;
 215 #endif
 216 }
 217
 218 int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
 219 int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
 220
 221 /*
 222  * Flags in backing_dev_info::capability
 223  *
 224  * The first three flags control whether dirty pages will contribute to the
 225  * VM's accounting and whether writepages() should be called for dirty pages
 226  * (something that would not, for example, be appropriate for ramfs)
 227  *
 228  * WARNING: these flags are closely related and should not normally be
 229  * used separately.  The BDI_CAP_NO_ACCT_AND_WRITEBACK combines these
 230  * three flags into a single convenience macro.
 231  *
 232  * BDI_CAP_NO_ACCT_DIRTY:  Dirty pages shouldn't contribute to accounting
 233  * BDI_CAP_NO_WRITEBACK:   Don't write pages back
 234  * BDI_CAP_NO_ACCT_WB:     Don't automatically account writeback pages
 235  *
 236  * These flags let !MMU mmap() govern direct device mapping vs immediate
 237  * copying more easily for MAP_PRIVATE, especially for ROM filesystems.
 238  *
 239  * BDI_CAP_MAP_COPY:       Copy can be mapped (MAP_PRIVATE)
 240  * BDI_CAP_MAP_DIRECT:     Can be mapped directly (MAP_SHARED)
 241  * BDI_CAP_READ_MAP:       Can be mapped for reading
 242  * BDI_CAP_WRITE_MAP:      Can be mapped for writing
 243  * BDI_CAP_EXEC_MAP:       Can be mapped for execution
 244  *
 245  * BDI_CAP_SWAP_BACKED:    Count shmem/tmpfs objects as swap-backed.
 246  *
 247  * BDI_CAP_STRICTLIMIT:    Keep number of dirty pages below bdi threshold.
 248  */
 249 #define BDI_CAP_NO_ACCT_DIRTY   0x00000001
 250 #define BDI_CAP_NO_WRITEBACK    0x00000002
 251 #define BDI_CAP_MAP_COPY        0x00000004
 252 #define BDI_CAP_MAP_DIRECT      0x00000008
 253 #define BDI_CAP_READ_MAP        0x00000010
 254 #define BDI_CAP_WRITE_MAP       0x00000020
 255 #define BDI_CAP_EXEC_MAP        0x00000040
 256 #define BDI_CAP_NO_ACCT_WB      0x00000080
 257 #define BDI_CAP_SWAP_BACKED     0x00000100
 258 #define BDI_CAP_STABLE_WRITES   0x00000200
 259 #define BDI_CAP_STRICTLIMIT     0x00000400
 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_stable_pages_required(struct backing_dev_info *bdi)
 315 {
 316         return bdi->capabilities & BDI_CAP_STABLE_WRITES;
 317 }
 318
 319 static inline bool bdi_cap_writeback_dirty(struct backing_dev_info *bdi)
 320 {
 321         return !(bdi->capabilities & BDI_CAP_NO_WRITEBACK);
 322 }
 323
 324 static inline bool bdi_cap_account_dirty(struct backing_dev_info *bdi)
 325 {
 326         return !(bdi->capabilities & BDI_CAP_NO_ACCT_DIRTY);
 327 }
 328
 329 static inline bool bdi_cap_account_writeback(struct backing_dev_info *bdi)
 330 {
 331         /* Paranoia: BDI_CAP_NO_WRITEBACK implies BDI_CAP_NO_ACCT_WB */
 332         return !(bdi->capabilities & (BDI_CAP_NO_ACCT_WB |
 333                                       BDI_CAP_NO_WRITEBACK));
 334 }
 335
 336 static inline bool bdi_cap_swap_backed(struct backing_dev_info *bdi)
 337 {
 338         return bdi->capabilities & BDI_CAP_SWAP_BACKED;
 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 */