1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __LINUX_BACKING_DEV_DEFS_H
3 #define __LINUX_BACKING_DEV_DEFS_H
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>
21 * Bits in bdi_writeback.state
24 WB_registered
, /* bdi_register() was done */
25 WB_writeback_running
, /* Writeback is in progress */
26 WB_has_dirty_io
, /* Dirty inodes on ->b_{dirty|io|more_io} */
27 WB_start_all
, /* nr_pages == 0 (all) work pending */
30 enum wb_congested_state
{
31 WB_async_congested
, /* The async (write) queue is getting full */
32 WB_sync_congested
, /* The sync queue is getting full */
35 typedef int (congested_fn
)(void *, int);
45 #define WB_STAT_BATCH (8*(1+ilog2(nr_cpu_ids)))
48 * why some writeback work was initiated
55 WB_REASON_LAPTOP_TIMER
,
56 WB_REASON_FREE_MORE_MEM
,
57 WB_REASON_FS_FREE_SPACE
,
59 * There is no bdi forker thread any more and works are done
60 * by emergency worker, however, this is TPs userland visible
61 * and we'll be exposing exactly the same information,
62 * so it has a mismatch name.
64 WB_REASON_FORKER_THREAD
,
70 * For cgroup writeback, multiple wb's may map to the same blkcg. Those
71 * wb's can operate mostly independently but should share the congested
72 * state. To facilitate such sharing, the congested state is tracked using
73 * the following struct which is created on demand, indexed by blkcg ID on
74 * its bdi, and refcounted.
76 struct bdi_writeback_congested
{
77 unsigned long state
; /* WB_[a]sync_congested flags */
78 atomic_t refcnt
; /* nr of attached wb's and blkg */
80 #ifdef CONFIG_CGROUP_WRITEBACK
81 struct backing_dev_info
*__bdi
; /* the associated bdi, set to NULL
82 * on bdi unregistration. For memcg-wb
83 * internal use only! */
84 int blkcg_id
; /* ID of the associated blkcg */
85 struct rb_node rb_node
; /* on bdi->cgwb_congestion_tree */
90 * Each wb (bdi_writeback) can perform writeback operations, is measured
91 * and throttled, independently. Without cgroup writeback, each bdi
92 * (bdi_writeback) is served by its embedded bdi->wb.
94 * On the default hierarchy, blkcg implicitly enables memcg. This allows
95 * using memcg's page ownership for attributing writeback IOs, and every
96 * memcg - blkcg combination can be served by its own wb by assigning a
97 * dedicated wb to each memcg, which enables isolation across different
98 * cgroups and propagation of IO back pressure down from the IO layer upto
99 * the tasks which are generating the dirty pages to be written back.
101 * A cgroup wb is indexed on its bdi by the ID of the associated memcg,
102 * refcounted with the number of inodes attached to it, and pins the memcg
103 * and the corresponding blkcg. As the corresponding blkcg for a memcg may
104 * change as blkcg is disabled and enabled higher up in the hierarchy, a wb
105 * is tested for blkcg after lookup and removed from index on mismatch so
106 * that a new wb for the combination can be created.
108 struct bdi_writeback
{
109 struct backing_dev_info
*bdi
; /* our parent bdi */
111 unsigned long state
; /* Always use atomic bitops on this */
112 unsigned long last_old_flush
; /* last old data flush */
114 struct list_head b_dirty
; /* dirty inodes */
115 struct list_head b_io
; /* parked for writeback */
116 struct list_head b_more_io
; /* parked for more writeback */
117 struct list_head b_dirty_time
; /* time stamps are dirty */
118 spinlock_t list_lock
; /* protects the b_* lists */
120 struct percpu_counter stat
[NR_WB_STAT_ITEMS
];
122 struct bdi_writeback_congested
*congested
;
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 */
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.
136 unsigned long dirty_ratelimit
;
137 unsigned long balanced_dirty_ratelimit
;
139 struct fprop_local_percpu completions
;
141 enum wb_reason start_all_reason
;
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 */
147 unsigned long dirty_sleep
; /* last wait */
149 struct list_head bdi_node
; /* anchored at bdi->wb_list */
151 #ifdef CONFIG_CGROUP_WRITEBACK
152 struct percpu_ref refcnt
; /* used only for !root wb's */
153 struct fprop_local_percpu memcg_completions
;
154 struct cgroup_subsys_state
*memcg_css
; /* the associated memcg */
155 struct cgroup_subsys_state
*blkcg_css
; /* and blkcg */
156 struct list_head memcg_node
; /* anchored at memcg->cgwb_list */
157 struct list_head blkcg_node
; /* anchored at blkcg->cgwb_list */
160 struct work_struct release_work
;
166 struct backing_dev_info
{
167 struct list_head bdi_list
;
168 unsigned long ra_pages
; /* max readahead in PAGE_SIZE units */
169 unsigned long io_pages
; /* max allowed IO size */
170 congested_fn
*congested_fn
; /* Function pointer if device is md/dm */
171 void *congested_data
; /* Pointer to aux data for congested func */
175 struct kref refcnt
; /* Reference counter for the structure */
176 unsigned int capabilities
; /* Device capabilities */
177 unsigned int min_ratio
;
178 unsigned int max_ratio
, max_prop_frac
;
181 * Sum of avg_write_bw of wbs with dirty inodes. > 0 if there are
182 * any dirty wbs, which is depended upon by bdi_has_dirty().
184 atomic_long_t tot_write_bandwidth
;
186 struct bdi_writeback wb
; /* the root writeback info for this bdi */
187 struct list_head wb_list
; /* list of all wbs */
188 #ifdef CONFIG_CGROUP_WRITEBACK
189 struct radix_tree_root cgwb_tree
; /* radix tree of active cgroup wbs */
190 struct rb_root cgwb_congested_tree
; /* their congested states */
191 struct mutex cgwb_release_mutex
; /* protect shutdown of wb structs */
192 struct rw_semaphore wb_switch_rwsem
; /* no cgwb switch while syncing */
194 struct bdi_writeback_congested
*wb_congested
;
196 wait_queue_head_t wb_waitq
;
199 struct device
*owner
;
201 struct timer_list laptop_mode_wb_timer
;
203 #ifdef CONFIG_DEBUG_FS
204 struct dentry
*debug_dir
;
205 struct dentry
*debug_stats
;
214 void clear_wb_congested(struct bdi_writeback_congested
*congested
, int sync
);
215 void set_wb_congested(struct bdi_writeback_congested
*congested
, int sync
);
217 static inline void clear_bdi_congested(struct backing_dev_info
*bdi
, int sync
)
219 clear_wb_congested(bdi
->wb
.congested
, sync
);
222 static inline void set_bdi_congested(struct backing_dev_info
*bdi
, int sync
)
224 set_wb_congested(bdi
->wb
.congested
, sync
);
227 struct wb_lock_cookie
{
232 #ifdef CONFIG_CGROUP_WRITEBACK
235 * wb_tryget - try to increment a wb's refcount
236 * @wb: bdi_writeback to get
238 static inline bool wb_tryget(struct bdi_writeback
*wb
)
240 if (wb
!= &wb
->bdi
->wb
)
241 return percpu_ref_tryget(&wb
->refcnt
);
246 * wb_get - increment a wb's refcount
247 * @wb: bdi_writeback to get
249 static inline void wb_get(struct bdi_writeback
*wb
)
251 if (wb
!= &wb
->bdi
->wb
)
252 percpu_ref_get(&wb
->refcnt
);
256 * wb_put - decrement a wb's refcount
257 * @wb: bdi_writeback to put
259 static inline void wb_put(struct bdi_writeback
*wb
)
261 if (WARN_ON_ONCE(!wb
->bdi
)) {
263 * A driver bug might cause a file to be removed before bdi was
269 if (wb
!= &wb
->bdi
->wb
)
270 percpu_ref_put(&wb
->refcnt
);
274 * wb_dying - is a wb dying?
275 * @wb: bdi_writeback of interest
277 * Returns whether @wb is unlinked and being drained.
279 static inline bool wb_dying(struct bdi_writeback
*wb
)
281 return percpu_ref_is_dying(&wb
->refcnt
);
284 #else /* CONFIG_CGROUP_WRITEBACK */
286 static inline bool wb_tryget(struct bdi_writeback
*wb
)
291 static inline void wb_get(struct bdi_writeback
*wb
)
295 static inline void wb_put(struct bdi_writeback
*wb
)
299 static inline bool wb_dying(struct bdi_writeback
*wb
)
304 #endif /* CONFIG_CGROUP_WRITEBACK */
306 #endif /* __LINUX_BACKING_DEV_DEFS_H */