include/linux/backing-dev-defs.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef __LINUX_BACKING_DEV_DEFS_H
   3 #define __LINUX_BACKING_DEV_DEFS_H
   4
   5 #include <linux/list.h>
   6 #include <linux/radix-tree.h>
   7 #include <linux/rbtree.h>
   8 #include <linux/spinlock.h>
   9 #include <linux/percpu_counter.h>
  10 #include <linux/percpu-refcount.h>
  11 #include <linux/flex_proportions.h>
  12 #include <linux/timer.h>
  13 #include <linux/workqueue.h>
  14 #include <linux/kref.h>
  15 #include <linux/refcount.h>
  16
  17 struct page;
  18 struct device;
  19 struct dentry;
  20
  21 /*
  22  * Bits in bdi_writeback.state
  23  */
  24 enum wb_state {
  25         WB_registered,          /* bdi_register() was done */
  26         WB_writeback_running,   /* Writeback is in progress */
  27         WB_has_dirty_io,        /* Dirty inodes on ->b_{dirty|io|more_io} */
  28         WB_start_all,           /* nr_pages == 0 (all) work pending */
  29 };
  30
  31 enum wb_congested_state {
  32         WB_async_congested,     /* The async (write) queue is getting full */
  33         WB_sync_congested,      /* The sync queue is getting full */
  34 };
  35
  36 enum wb_stat_item {
  37         WB_RECLAIMABLE,
  38         WB_WRITEBACK,
  39         WB_DIRTIED,
  40         WB_WRITTEN,
  41         NR_WB_STAT_ITEMS
  42 };
  43
  44 #define WB_STAT_BATCH (8*(1+ilog2(nr_cpu_ids)))
  45
  46 /*
  47  * why some writeback work was initiated
  48  */
  49 enum wb_reason {
  50         WB_REASON_BACKGROUND,
  51         WB_REASON_VMSCAN,
  52         WB_REASON_SYNC,
  53         WB_REASON_PERIODIC,
  54         WB_REASON_LAPTOP_TIMER,
  55         WB_REASON_FS_FREE_SPACE,
  56         /*
  57          * There is no bdi forker thread any more and works are done
  58          * by emergency worker, however, this is TPs userland visible
  59          * and we'll be exposing exactly the same information,
  60          * so it has a mismatch name.
  61          */
  62         WB_REASON_FORKER_THREAD,
  63         WB_REASON_FOREIGN_FLUSH,
  64
  65         WB_REASON_MAX,
  66 };
  67
  68 struct wb_completion {
  69         atomic_t                cnt;
  70         wait_queue_head_t       *waitq;
  71 };
  72
  73 #define __WB_COMPLETION_INIT(_waitq)    \
  74         (struct wb_completion){ .cnt = ATOMIC_INIT(1), .waitq = (_waitq) }
  75
  76 /*
  77  * If one wants to wait for one or more wb_writeback_works, each work's
  78  * ->done should be set to a wb_completion defined using the following
  79  * macro.  Once all work items are issued with wb_queue_work(), the caller
  80  * can wait for the completion of all using wb_wait_for_completion().  Work
  81  * items which are waited upon aren't freed automatically on completion.
  82  */
  83 #define WB_COMPLETION_INIT(bdi)         __WB_COMPLETION_INIT(&(bdi)->wb_waitq)
  84
  85 #define DEFINE_WB_COMPLETION(cmpl, bdi) \
  86         struct wb_completion cmpl = WB_COMPLETION_INIT(bdi)
  87
  88 /*
  89  * Each wb (bdi_writeback) can perform writeback operations, is measured
  90  * and throttled, independently.  Without cgroup writeback, each bdi
  91  * (bdi_writeback) is served by its embedded bdi->wb.
  92  *
  93  * On the default hierarchy, blkcg implicitly enables memcg.  This allows
  94  * using memcg's page ownership for attributing writeback IOs, and every
  95  * memcg - blkcg combination can be served by its own wb by assigning a
  96  * dedicated wb to each memcg, which enables isolation across different
  97  * cgroups and propagation of IO back pressure down from the IO layer upto
  98  * the tasks which are generating the dirty pages to be written back.
  99  *
 100  * A cgroup wb is indexed on its bdi by the ID of the associated memcg,
 101  * refcounted with the number of inodes attached to it, and pins the memcg
 102  * and the corresponding blkcg.  As the corresponding blkcg for a memcg may
 103  * change as blkcg is disabled and enabled higher up in the hierarchy, a wb
 104  * is tested for blkcg after lookup and removed from index on mismatch so
 105  * that a new wb for the combination can be created.
 106  */
 107 struct bdi_writeback {
 108         struct backing_dev_info *bdi;   /* our parent bdi */
 109
 110         unsigned long state;            /* Always use atomic bitops on this */
 111         unsigned long last_old_flush;   /* last old data flush */
 112
 113         struct list_head b_dirty;       /* dirty inodes */
 114         struct list_head b_io;          /* parked for writeback */
 115         struct list_head b_more_io;     /* parked for more writeback */
 116         struct list_head b_dirty_time;  /* time stamps are dirty */
 117         spinlock_t list_lock;           /* protects the b_* lists */
 118
 119         atomic_t writeback_inodes;      /* number of inodes under writeback */
 120         struct percpu_counter stat[NR_WB_STAT_ITEMS];
 121
 122         unsigned long congested;        /* WB_[a]sync_congested flags */
 123
 124         unsigned long bw_time_stamp;    /* last time write bw is updated */
 125         unsigned long dirtied_stamp;
 126         unsigned long written_stamp;    /* pages written at bw_time_stamp */
 127         unsigned long write_bandwidth;  /* the estimated write bandwidth */
 128         unsigned long avg_write_bandwidth; /* further smoothed write bw, > 0 */
 129
 130         /*
 131          * The base dirty throttle rate, re-calculated on every 200ms.
 132          * All the bdi tasks' dirty rate will be curbed under it.
 133          * @dirty_ratelimit tracks the estimated @balanced_dirty_ratelimit
 134          * in small steps and is much more smooth/stable than the latter.
 135          */
 136         unsigned long dirty_ratelimit;
 137         unsigned long balanced_dirty_ratelimit;
 138
 139         struct fprop_local_percpu completions;
 140         int dirty_exceeded;
 141         enum wb_reason start_all_reason;
 142
 143         spinlock_t work_lock;           /* protects work_list & dwork scheduling */
 144         struct list_head work_list;
 145         struct delayed_work dwork;      /* work item used for writeback */
 146         struct delayed_work bw_dwork;   /* work item used for bandwidth estimate */
 147
 148         unsigned long dirty_sleep;      /* last wait */
 149
 150         struct list_head bdi_node;      /* anchored at bdi->wb_list */
 151
 152 #ifdef CONFIG_CGROUP_WRITEBACK
 153         struct percpu_ref refcnt;       /* used only for !root wb's */
 154         struct fprop_local_percpu memcg_completions;
 155         struct cgroup_subsys_state *memcg_css; /* the associated memcg */
 156         struct cgroup_subsys_state *blkcg_css; /* and blkcg */
 157         struct list_head memcg_node;    /* anchored at memcg->cgwb_list */
 158         struct list_head blkcg_node;    /* anchored at blkcg->cgwb_list */
 159         struct list_head b_attached;    /* attached inodes, protected by list_lock */
 160         struct list_head offline_node;  /* anchored at offline_cgwbs */
 161
 162         union {
 163                 struct work_struct release_work;
 164                 struct rcu_head rcu;
 165         };
 166 #endif
 167 };
 168
 169 struct backing_dev_info {
 170         u64 id;
 171         struct rb_node rb_node; /* keyed by ->id */
 172         struct list_head bdi_list;
 173         unsigned long ra_pages; /* max readahead in PAGE_SIZE units */
 174         unsigned long io_pages; /* max allowed IO size */
 175
 176         struct kref refcnt;     /* Reference counter for the structure */
 177         unsigned int capabilities; /* Device capabilities */
 178         unsigned int min_ratio;
 179         unsigned int max_ratio, max_prop_frac;
 180
 181         /*
 182          * Sum of avg_write_bw of wbs with dirty inodes.  > 0 if there are
 183          * any dirty wbs, which is depended upon by bdi_has_dirty().
 184          */
 185         atomic_long_t tot_write_bandwidth;
 186
 187         struct bdi_writeback wb;  /* the root writeback info for this bdi */
 188         struct list_head wb_list; /* list of all wbs */
 189 #ifdef CONFIG_CGROUP_WRITEBACK
 190         struct radix_tree_root cgwb_tree; /* radix tree of active cgroup wbs */
 191         struct mutex cgwb_release_mutex;  /* protect shutdown of wb structs */
 192         struct rw_semaphore wb_switch_rwsem; /* no cgwb switch while syncing */
 193 #endif
 194         wait_queue_head_t wb_waitq;
 195
 196         struct device *dev;
 197         char dev_name[64];
 198         struct device *owner;
 199
 200         struct timer_list laptop_mode_wb_timer;
 201
 202 #ifdef CONFIG_DEBUG_FS
 203         struct dentry *debug_dir;
 204 #endif
 205 };
 206
 207 enum {
 208         BLK_RW_ASYNC    = 0,
 209         BLK_RW_SYNC     = 1,
 210 };
 211
 212 void clear_bdi_congested(struct backing_dev_info *bdi, int sync);
 213 void set_bdi_congested(struct backing_dev_info *bdi, int sync);
 214
 215 struct wb_lock_cookie {
 216         bool locked;
 217         unsigned long flags;
 218 };
 219
 220 #ifdef CONFIG_CGROUP_WRITEBACK
 221
 222 /**
 223  * wb_tryget - try to increment a wb's refcount
 224  * @wb: bdi_writeback to get
 225  */
 226 static inline bool wb_tryget(struct bdi_writeback *wb)
 227 {
 228         if (wb != &wb->bdi->wb)
 229                 return percpu_ref_tryget(&wb->refcnt);
 230         return true;
 231 }
 232
 233 /**
 234  * wb_get - increment a wb's refcount
 235  * @wb: bdi_writeback to get
 236  */
 237 static inline void wb_get(struct bdi_writeback *wb)
 238 {
 239         if (wb != &wb->bdi->wb)
 240                 percpu_ref_get(&wb->refcnt);
 241 }
 242
 243 /**
 244  * wb_put - decrement a wb's refcount
 245  * @wb: bdi_writeback to put
 246  * @nr: number of references to put
 247  */
 248 static inline void wb_put_many(struct bdi_writeback *wb, unsigned long nr)
 249 {
 250         if (WARN_ON_ONCE(!wb->bdi)) {
 251                 /*
 252                  * A driver bug might cause a file to be removed before bdi was
 253                  * initialized.
 254                  */
 255                 return;
 256         }
 257
 258         if (wb != &wb->bdi->wb)
 259                 percpu_ref_put_many(&wb->refcnt, nr);
 260 }
 261
 262 /**
 263  * wb_put - decrement a wb's refcount
 264  * @wb: bdi_writeback to put
 265  */
 266 static inline void wb_put(struct bdi_writeback *wb)
 267 {
 268         wb_put_many(wb, 1);
 269 }
 270
 271 /**
 272  * wb_dying - is a wb dying?
 273  * @wb: bdi_writeback of interest
 274  *
 275  * Returns whether @wb is unlinked and being drained.
 276  */
 277 static inline bool wb_dying(struct bdi_writeback *wb)
 278 {
 279         return percpu_ref_is_dying(&wb->refcnt);
 280 }
 281
 282 #else   /* CONFIG_CGROUP_WRITEBACK */
 283
 284 static inline bool wb_tryget(struct bdi_writeback *wb)
 285 {
 286         return true;
 287 }
 288
 289 static inline void wb_get(struct bdi_writeback *wb)
 290 {
 291 }
 292
 293 static inline void wb_put(struct bdi_writeback *wb)
 294 {
 295 }
 296
 297 static inline void wb_put_many(struct bdi_writeback *wb, unsigned long nr)
 298 {
 299 }
 300
 301 static inline bool wb_dying(struct bdi_writeback *wb)
 302 {
 303         return false;
 304 }
 305
 306 #endif  /* CONFIG_CGROUP_WRITEBACK */
 307
 308 #endif  /* __LINUX_BACKING_DEV_DEFS_H */