include/linux/backing-dev-defs.h

   1 #ifndef __LINUX_BACKING_DEV_DEFS_H
   2 #define __LINUX_BACKING_DEV_DEFS_H
   3
   4 #include <linux/list.h>
   5 #include <linux/radix-tree.h>
   6 #include <linux/rbtree.h>
   7 #include <linux/spinlock.h>
   8 #include <linux/percpu_counter.h>
   9 #include <linux/percpu-refcount.h>
  10 #include <linux/flex_proportions.h>
  11 #include <linux/timer.h>
  12 #include <linux/workqueue.h>
  13
  14 struct page;
  15 struct device;
  16 struct dentry;
  17
  18 /*
  19  * Bits in bdi_writeback.state
  20  */
  21 enum wb_state {
  22         WB_registered,          /* bdi_register() was done */
  23         WB_writeback_running,   /* Writeback is in progress */
  24         WB_has_dirty_io,        /* Dirty inodes on ->b_{dirty|io|more_io} */
  25 };
  26
  27 enum wb_congested_state {
  28         WB_async_congested,     /* The async (write) queue is getting full */
  29         WB_sync_congested,      /* The sync queue is getting full */
  30 };
  31
  32 typedef int (congested_fn)(void *, int);
  33
  34 enum wb_stat_item {
  35         WB_RECLAIMABLE,
  36         WB_WRITEBACK,
  37         WB_DIRTIED,
  38         WB_WRITTEN,
  39         NR_WB_STAT_ITEMS
  40 };
  41
  42 #define WB_STAT_BATCH (8*(1+ilog2(nr_cpu_ids)))
  43
  44 /*
  45  * For cgroup writeback, multiple wb's may map to the same blkcg.  Those
  46  * wb's can operate mostly independently but should share the congested
  47  * state.  To facilitate such sharing, the congested state is tracked using
  48  * the following struct which is created on demand, indexed by blkcg ID on
  49  * its bdi, and refcounted.
  50  */
  51 struct bdi_writeback_congested {
  52         unsigned long state;            /* WB_[a]sync_congested flags */
  53         atomic_t refcnt;                /* nr of attached wb's and blkg */
  54
  55 #ifdef CONFIG_CGROUP_WRITEBACK
  56         struct backing_dev_info *bdi;   /* the associated bdi */
  57         int blkcg_id;                   /* ID of the associated blkcg */
  58         struct rb_node rb_node;         /* on bdi->cgwb_congestion_tree */
  59 #endif
  60 };
  61
  62 /*
  63  * Each wb (bdi_writeback) can perform writeback operations, is measured
  64  * and throttled, independently.  Without cgroup writeback, each bdi
  65  * (bdi_writeback) is served by its embedded bdi->wb.
  66  *
  67  * On the default hierarchy, blkcg implicitly enables memcg.  This allows
  68  * using memcg's page ownership for attributing writeback IOs, and every
  69  * memcg - blkcg combination can be served by its own wb by assigning a
  70  * dedicated wb to each memcg, which enables isolation across different
  71  * cgroups and propagation of IO back pressure down from the IO layer upto
  72  * the tasks which are generating the dirty pages to be written back.
  73  *
  74  * A cgroup wb is indexed on its bdi by the ID of the associated memcg,
  75  * refcounted with the number of inodes attached to it, and pins the memcg
  76  * and the corresponding blkcg.  As the corresponding blkcg for a memcg may
  77  * change as blkcg is disabled and enabled higher up in the hierarchy, a wb
  78  * is tested for blkcg after lookup and removed from index on mismatch so
  79  * that a new wb for the combination can be created.
  80  */
  81 struct bdi_writeback {
  82         struct backing_dev_info *bdi;   /* our parent bdi */
  83
  84         unsigned long state;            /* Always use atomic bitops on this */
  85         unsigned long last_old_flush;   /* last old data flush */
  86
  87         struct list_head b_dirty;       /* dirty inodes */
  88         struct list_head b_io;          /* parked for writeback */
  89         struct list_head b_more_io;     /* parked for more writeback */
  90         struct list_head b_dirty_time;  /* time stamps are dirty */
  91         spinlock_t list_lock;           /* protects the b_* lists */
  92
  93         struct percpu_counter stat[NR_WB_STAT_ITEMS];
  94
  95         struct bdi_writeback_congested *congested;
  96
  97         unsigned long bw_time_stamp;    /* last time write bw is updated */
  98         unsigned long dirtied_stamp;
  99         unsigned long written_stamp;    /* pages written at bw_time_stamp */
 100         unsigned long write_bandwidth;  /* the estimated write bandwidth */
 101         unsigned long avg_write_bandwidth; /* further smoothed write bw, > 0 */
 102
 103         /*
 104          * The base dirty throttle rate, re-calculated on every 200ms.
 105          * All the bdi tasks' dirty rate will be curbed under it.
 106          * @dirty_ratelimit tracks the estimated @balanced_dirty_ratelimit
 107          * in small steps and is much more smooth/stable than the latter.
 108          */
 109         unsigned long dirty_ratelimit;
 110         unsigned long balanced_dirty_ratelimit;
 111
 112         struct fprop_local_percpu completions;
 113         int dirty_exceeded;
 114
 115         spinlock_t work_lock;           /* protects work_list & dwork scheduling */
 116         struct list_head work_list;
 117         struct delayed_work dwork;      /* work item used for writeback */
 118
 119         unsigned long dirty_sleep;      /* last wait */
 120
 121         struct list_head bdi_node;      /* anchored at bdi->wb_list */
 122
 123 #ifdef CONFIG_CGROUP_WRITEBACK
 124         struct percpu_ref refcnt;       /* used only for !root wb's */
 125         struct fprop_local_percpu memcg_completions;
 126         struct cgroup_subsys_state *memcg_css; /* the associated memcg */
 127         struct cgroup_subsys_state *blkcg_css; /* and blkcg */
 128         struct list_head memcg_node;    /* anchored at memcg->cgwb_list */
 129         struct list_head blkcg_node;    /* anchored at blkcg->cgwb_list */
 130
 131         union {
 132                 struct work_struct release_work;
 133                 struct rcu_head rcu;
 134         };
 135 #endif
 136 };
 137
 138 struct backing_dev_info {
 139         struct list_head bdi_list;
 140         unsigned long ra_pages; /* max readahead in PAGE_SIZE units */
 141         unsigned long io_pages; /* max allowed IO size */
 142         congested_fn *congested_fn; /* Function pointer if device is md/dm */
 143         void *congested_data;   /* Pointer to aux data for congested func */
 144
 145         char *name;
 146
 147         unsigned int capabilities; /* Device capabilities */
 148         unsigned int min_ratio;
 149         unsigned int max_ratio, max_prop_frac;
 150
 151         /*
 152          * Sum of avg_write_bw of wbs with dirty inodes.  > 0 if there are
 153          * any dirty wbs, which is depended upon by bdi_has_dirty().
 154          */
 155         atomic_long_t tot_write_bandwidth;
 156
 157         struct bdi_writeback wb;  /* the root writeback info for this bdi */
 158         struct list_head wb_list; /* list of all wbs */
 159 #ifdef CONFIG_CGROUP_WRITEBACK
 160         struct radix_tree_root cgwb_tree; /* radix tree of active cgroup wbs */
 161         struct rb_root cgwb_congested_tree; /* their congested states */
 162         atomic_t usage_cnt; /* counts both cgwbs and cgwb_contested's */
 163 #else
 164         struct bdi_writeback_congested *wb_congested;
 165 #endif
 166         wait_queue_head_t wb_waitq;
 167
 168         struct device *dev;
 169         struct device *owner;
 170
 171         struct timer_list laptop_mode_wb_timer;
 172
 173 #ifdef CONFIG_DEBUG_FS
 174         struct dentry *debug_dir;
 175         struct dentry *debug_stats;
 176 #endif
 177 };
 178
 179 enum {
 180         BLK_RW_ASYNC    = 0,
 181         BLK_RW_SYNC     = 1,
 182 };
 183
 184 void clear_wb_congested(struct bdi_writeback_congested *congested, int sync);
 185 void set_wb_congested(struct bdi_writeback_congested *congested, int sync);
 186
 187 static inline void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
 188 {
 189         clear_wb_congested(bdi->wb.congested, sync);
 190 }
 191
 192 static inline void set_bdi_congested(struct backing_dev_info *bdi, int sync)
 193 {
 194         set_wb_congested(bdi->wb.congested, sync);
 195 }
 196
 197 #ifdef CONFIG_CGROUP_WRITEBACK
 198
 199 /**
 200  * wb_tryget - try to increment a wb's refcount
 201  * @wb: bdi_writeback to get
 202  */
 203 static inline bool wb_tryget(struct bdi_writeback *wb)
 204 {
 205         if (wb != &wb->bdi->wb)
 206                 return percpu_ref_tryget(&wb->refcnt);
 207         return true;
 208 }
 209
 210 /**
 211  * wb_get - increment a wb's refcount
 212  * @wb: bdi_writeback to get
 213  */
 214 static inline void wb_get(struct bdi_writeback *wb)
 215 {
 216         if (wb != &wb->bdi->wb)
 217                 percpu_ref_get(&wb->refcnt);
 218 }
 219
 220 /**
 221  * wb_put - decrement a wb's refcount
 222  * @wb: bdi_writeback to put
 223  */
 224 static inline void wb_put(struct bdi_writeback *wb)
 225 {
 226         if (wb != &wb->bdi->wb)
 227                 percpu_ref_put(&wb->refcnt);
 228 }
 229
 230 /**
 231  * wb_dying - is a wb dying?
 232  * @wb: bdi_writeback of interest
 233  *
 234  * Returns whether @wb is unlinked and being drained.
 235  */
 236 static inline bool wb_dying(struct bdi_writeback *wb)
 237 {
 238         return percpu_ref_is_dying(&wb->refcnt);
 239 }
 240
 241 #else   /* CONFIG_CGROUP_WRITEBACK */
 242
 243 static inline bool wb_tryget(struct bdi_writeback *wb)
 244 {
 245         return true;
 246 }
 247
 248 static inline void wb_get(struct bdi_writeback *wb)
 249 {
 250 }
 251
 252 static inline void wb_put(struct bdi_writeback *wb)
 253 {
 254 }
 255
 256 static inline bool wb_dying(struct bdi_writeback *wb)
 257 {
 258         return false;
 259 }
 260
 261 #endif  /* CONFIG_CGROUP_WRITEBACK */
 262
 263 #endif  /* __LINUX_BACKING_DEV_DEFS_H */