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 #include <linux/kref.h>
  14
  15 struct page;
  16 struct device;
  17 struct dentry;
  18
  19 /*
  20  * Bits in bdi_writeback.state
  21  */
  22 enum wb_state {
  23         WB_registered,          /* bdi_register() was done */
  24         WB_writeback_running,   /* Writeback is in progress */
  25         WB_has_dirty_io,        /* Dirty inodes on ->b_{dirty|io|more_io} */
  26 };
  27
  28 enum wb_congested_state {
  29         WB_async_congested,     /* The async (write) queue is getting full */
  30         WB_sync_congested,      /* The sync queue is getting full */
  31 };
  32
  33 typedef int (congested_fn)(void *, int);
  34
  35 enum wb_stat_item {
  36         WB_RECLAIMABLE,
  37         WB_WRITEBACK,
  38         WB_DIRTIED,
  39         WB_WRITTEN,
  40         NR_WB_STAT_ITEMS
  41 };
  42
  43 #define WB_STAT_BATCH (8*(1+ilog2(nr_cpu_ids)))
  44
  45 /*
  46  * For cgroup writeback, multiple wb's may map to the same blkcg.  Those
  47  * wb's can operate mostly independently but should share the congested
  48  * state.  To facilitate such sharing, the congested state is tracked using
  49  * the following struct which is created on demand, indexed by blkcg ID on
  50  * its bdi, and refcounted.
  51  */
  52 struct bdi_writeback_congested {
  53         unsigned long state;            /* WB_[a]sync_congested flags */
  54         atomic_t refcnt;                /* nr of attached wb's and blkg */
  55
  56 #ifdef CONFIG_CGROUP_WRITEBACK
  57         struct backing_dev_info *bdi;   /* the associated bdi */
  58         int blkcg_id;                   /* ID of the associated blkcg */
  59         struct rb_node rb_node;         /* on bdi->cgwb_congestion_tree */
  60 #endif
  61 };
  62
  63 /*
  64  * Each wb (bdi_writeback) can perform writeback operations, is measured
  65  * and throttled, independently.  Without cgroup writeback, each bdi
  66  * (bdi_writeback) is served by its embedded bdi->wb.
  67  *
  68  * On the default hierarchy, blkcg implicitly enables memcg.  This allows
  69  * using memcg's page ownership for attributing writeback IOs, and every
  70  * memcg - blkcg combination can be served by its own wb by assigning a
  71  * dedicated wb to each memcg, which enables isolation across different
  72  * cgroups and propagation of IO back pressure down from the IO layer upto
  73  * the tasks which are generating the dirty pages to be written back.
  74  *
  75  * A cgroup wb is indexed on its bdi by the ID of the associated memcg,
  76  * refcounted with the number of inodes attached to it, and pins the memcg
  77  * and the corresponding blkcg.  As the corresponding blkcg for a memcg may
  78  * change as blkcg is disabled and enabled higher up in the hierarchy, a wb
  79  * is tested for blkcg after lookup and removed from index on mismatch so
  80  * that a new wb for the combination can be created.
  81  */
  82 struct bdi_writeback {
  83         struct backing_dev_info *bdi;   /* our parent bdi */
  84
  85         unsigned long state;            /* Always use atomic bitops on this */
  86         unsigned long last_old_flush;   /* last old data flush */
  87
  88         struct list_head b_dirty;       /* dirty inodes */
  89         struct list_head b_io;          /* parked for writeback */
  90         struct list_head b_more_io;     /* parked for more writeback */
  91         struct list_head b_dirty_time;  /* time stamps are dirty */
  92         spinlock_t list_lock;           /* protects the b_* lists */
  93
  94         struct percpu_counter stat[NR_WB_STAT_ITEMS];
  95
  96         struct bdi_writeback_congested *congested;
  97
  98         unsigned long bw_time_stamp;    /* last time write bw is updated */
  99         unsigned long dirtied_stamp;
 100         unsigned long written_stamp;    /* pages written at bw_time_stamp */
 101         unsigned long write_bandwidth;  /* the estimated write bandwidth */
 102         unsigned long avg_write_bandwidth; /* further smoothed write bw, > 0 */
 103
 104         /*
 105          * The base dirty throttle rate, re-calculated on every 200ms.
 106          * All the bdi tasks' dirty rate will be curbed under it.
 107          * @dirty_ratelimit tracks the estimated @balanced_dirty_ratelimit
 108          * in small steps and is much more smooth/stable than the latter.
 109          */
 110         unsigned long dirty_ratelimit;
 111         unsigned long balanced_dirty_ratelimit;
 112
 113         struct fprop_local_percpu completions;
 114         int dirty_exceeded;
 115
 116         spinlock_t work_lock;           /* protects work_list & dwork scheduling */
 117         struct list_head work_list;
 118         struct delayed_work dwork;      /* work item used for writeback */
 119
 120         unsigned long dirty_sleep;      /* last wait */
 121
 122         struct list_head bdi_node;      /* anchored at bdi->wb_list */
 123
 124 #ifdef CONFIG_CGROUP_WRITEBACK
 125         struct percpu_ref refcnt;       /* used only for !root wb's */
 126         struct fprop_local_percpu memcg_completions;
 127         struct cgroup_subsys_state *memcg_css; /* the associated memcg */
 128         struct cgroup_subsys_state *blkcg_css; /* and blkcg */
 129         struct list_head memcg_node;    /* anchored at memcg->cgwb_list */
 130         struct list_head blkcg_node;    /* anchored at blkcg->cgwb_list */
 131
 132         union {
 133                 struct work_struct release_work;
 134                 struct rcu_head rcu;
 135         };
 136 #endif
 137 };
 138
 139 struct backing_dev_info {
 140         struct list_head bdi_list;
 141         unsigned long ra_pages; /* max readahead in PAGE_SIZE units */
 142         unsigned long io_pages; /* max allowed IO size */
 143         congested_fn *congested_fn; /* Function pointer if device is md/dm */
 144         void *congested_data;   /* Pointer to aux data for congested func */
 145
 146         char *name;
 147
 148         struct kref refcnt;     /* Reference counter for the structure */
 149         unsigned int capabilities; /* Device capabilities */
 150         unsigned int min_ratio;
 151         unsigned int max_ratio, max_prop_frac;
 152
 153         /*
 154          * Sum of avg_write_bw of wbs with dirty inodes.  > 0 if there are
 155          * any dirty wbs, which is depended upon by bdi_has_dirty().
 156          */
 157         atomic_long_t tot_write_bandwidth;
 158
 159         struct bdi_writeback wb;  /* the root writeback info for this bdi */
 160         struct list_head wb_list; /* list of all wbs */
 161 #ifdef CONFIG_CGROUP_WRITEBACK
 162         struct radix_tree_root cgwb_tree; /* radix tree of active cgroup wbs */
 163         struct rb_root cgwb_congested_tree; /* their congested states */
 164         atomic_t usage_cnt; /* counts both cgwbs and cgwb_contested's */
 165 #else
 166         struct bdi_writeback_congested *wb_congested;
 167 #endif
 168         wait_queue_head_t wb_waitq;
 169
 170         struct device *dev;
 171         struct device *owner;
 172
 173         struct timer_list laptop_mode_wb_timer;
 174
 175 #ifdef CONFIG_DEBUG_FS
 176         struct dentry *debug_dir;
 177         struct dentry *debug_stats;
 178 #endif
 179 };
 180
 181 enum {
 182         BLK_RW_ASYNC    = 0,
 183         BLK_RW_SYNC     = 1,
 184 };
 185
 186 void clear_wb_congested(struct bdi_writeback_congested *congested, int sync);
 187 void set_wb_congested(struct bdi_writeback_congested *congested, int sync);
 188
 189 static inline void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
 190 {
 191         clear_wb_congested(bdi->wb.congested, sync);
 192 }
 193
 194 static inline void set_bdi_congested(struct backing_dev_info *bdi, int sync)
 195 {
 196         set_wb_congested(bdi->wb.congested, sync);
 197 }
 198
 199 #ifdef CONFIG_CGROUP_WRITEBACK
 200
 201 /**
 202  * wb_tryget - try to increment a wb's refcount
 203  * @wb: bdi_writeback to get
 204  */
 205 static inline bool wb_tryget(struct bdi_writeback *wb)
 206 {
 207         if (wb != &wb->bdi->wb)
 208                 return percpu_ref_tryget(&wb->refcnt);
 209         return true;
 210 }
 211
 212 /**
 213  * wb_get - increment a wb's refcount
 214  * @wb: bdi_writeback to get
 215  */
 216 static inline void wb_get(struct bdi_writeback *wb)
 217 {
 218         if (wb != &wb->bdi->wb)
 219                 percpu_ref_get(&wb->refcnt);
 220 }
 221
 222 /**
 223  * wb_put - decrement a wb's refcount
 224  * @wb: bdi_writeback to put
 225  */
 226 static inline void wb_put(struct bdi_writeback *wb)
 227 {
 228         if (wb != &wb->bdi->wb)
 229                 percpu_ref_put(&wb->refcnt);
 230 }
 231
 232 /**
 233  * wb_dying - is a wb dying?
 234  * @wb: bdi_writeback of interest
 235  *
 236  * Returns whether @wb is unlinked and being drained.
 237  */
 238 static inline bool wb_dying(struct bdi_writeback *wb)
 239 {
 240         return percpu_ref_is_dying(&wb->refcnt);
 241 }
 242
 243 #else   /* CONFIG_CGROUP_WRITEBACK */
 244
 245 static inline bool wb_tryget(struct bdi_writeback *wb)
 246 {
 247         return true;
 248 }
 249
 250 static inline void wb_get(struct bdi_writeback *wb)
 251 {
 252 }
 253
 254 static inline void wb_put(struct bdi_writeback *wb)
 255 {
 256 }
 257
 258 static inline bool wb_dying(struct bdi_writeback *wb)
 259 {
 260         return false;
 261 }
 262
 263 #endif  /* CONFIG_CGROUP_WRITEBACK */
 264
 265 #endif  /* __LINUX_BACKING_DEV_DEFS_H */