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_shutting_down,	/* wb_shutdown() in progress */
25 	WB_writeback_running,	/* Writeback is in progress */
26 	WB_has_dirty_io,	/* Dirty inodes on ->b_{dirty|io|more_io} */
27 };
28 
29 enum wb_congested_state {
30 	WB_async_congested,	/* The async (write) queue is getting full */
31 	WB_sync_congested,	/* The sync queue is getting full */
32 };
33 
34 typedef int (congested_fn)(void *, int);
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  * For cgroup writeback, multiple wb's may map to the same blkcg.  Those
48  * wb's can operate mostly independently but should share the congested
49  * state.  To facilitate such sharing, the congested state is tracked using
50  * the following struct which is created on demand, indexed by blkcg ID on
51  * its bdi, and refcounted.
52  */
53 struct bdi_writeback_congested {
54 	unsigned long state;		/* WB_[a]sync_congested flags */
55 	atomic_t refcnt;		/* nr of attached wb's and blkg */
56 
57 #ifdef CONFIG_CGROUP_WRITEBACK
58 	struct backing_dev_info *__bdi;	/* the associated bdi, set to NULL
59 					 * on bdi unregistration. For memcg-wb
60 					 * internal use only! */
61 	int blkcg_id;			/* ID of the associated blkcg */
62 	struct rb_node rb_node;		/* on bdi->cgwb_congestion_tree */
63 #endif
64 };
65 
66 /*
67  * Each wb (bdi_writeback) can perform writeback operations, is measured
68  * and throttled, independently.  Without cgroup writeback, each bdi
69  * (bdi_writeback) is served by its embedded bdi->wb.
70  *
71  * On the default hierarchy, blkcg implicitly enables memcg.  This allows
72  * using memcg's page ownership for attributing writeback IOs, and every
73  * memcg - blkcg combination can be served by its own wb by assigning a
74  * dedicated wb to each memcg, which enables isolation across different
75  * cgroups and propagation of IO back pressure down from the IO layer upto
76  * the tasks which are generating the dirty pages to be written back.
77  *
78  * A cgroup wb is indexed on its bdi by the ID of the associated memcg,
79  * refcounted with the number of inodes attached to it, and pins the memcg
80  * and the corresponding blkcg.  As the corresponding blkcg for a memcg may
81  * change as blkcg is disabled and enabled higher up in the hierarchy, a wb
82  * is tested for blkcg after lookup and removed from index on mismatch so
83  * that a new wb for the combination can be created.
84  */
85 struct bdi_writeback {
86 	struct backing_dev_info *bdi;	/* our parent bdi */
87 
88 	unsigned long state;		/* Always use atomic bitops on this */
89 	unsigned long last_old_flush;	/* last old data flush */
90 
91 	struct list_head b_dirty;	/* dirty inodes */
92 	struct list_head b_io;		/* parked for writeback */
93 	struct list_head b_more_io;	/* parked for more writeback */
94 	struct list_head b_dirty_time;	/* time stamps are dirty */
95 	spinlock_t list_lock;		/* protects the b_* lists */
96 
97 	struct percpu_counter stat[NR_WB_STAT_ITEMS];
98 
99 	struct bdi_writeback_congested *congested;
100 
101 	unsigned long bw_time_stamp;	/* last time write bw is updated */
102 	unsigned long dirtied_stamp;
103 	unsigned long written_stamp;	/* pages written at bw_time_stamp */
104 	unsigned long write_bandwidth;	/* the estimated write bandwidth */
105 	unsigned long avg_write_bandwidth; /* further smoothed write bw, > 0 */
106 
107 	/*
108 	 * The base dirty throttle rate, re-calculated on every 200ms.
109 	 * All the bdi tasks' dirty rate will be curbed under it.
110 	 * @dirty_ratelimit tracks the estimated @balanced_dirty_ratelimit
111 	 * in small steps and is much more smooth/stable than the latter.
112 	 */
113 	unsigned long dirty_ratelimit;
114 	unsigned long balanced_dirty_ratelimit;
115 
116 	struct fprop_local_percpu completions;
117 	int dirty_exceeded;
118 
119 	spinlock_t work_lock;		/* protects work_list & dwork scheduling */
120 	struct list_head work_list;
121 	struct delayed_work dwork;	/* work item used for writeback */
122 
123 	unsigned long dirty_sleep;	/* last wait */
124 
125 	struct list_head bdi_node;	/* anchored at bdi->wb_list */
126 
127 #ifdef CONFIG_CGROUP_WRITEBACK
128 	struct percpu_ref refcnt;	/* used only for !root wb's */
129 	struct fprop_local_percpu memcg_completions;
130 	struct cgroup_subsys_state *memcg_css; /* the associated memcg */
131 	struct cgroup_subsys_state *blkcg_css; /* and blkcg */
132 	struct list_head memcg_node;	/* anchored at memcg->cgwb_list */
133 	struct list_head blkcg_node;	/* anchored at blkcg->cgwb_list */
134 
135 	union {
136 		struct work_struct release_work;
137 		struct rcu_head rcu;
138 	};
139 #endif
140 };
141 
142 struct backing_dev_info {
143 	struct list_head bdi_list;
144 	unsigned long ra_pages;	/* max readahead in PAGE_SIZE units */
145 	unsigned long io_pages;	/* max allowed IO size */
146 	congested_fn *congested_fn; /* Function pointer if device is md/dm */
147 	void *congested_data;	/* Pointer to aux data for congested func */
148 
149 	const char *name;
150 
151 	struct kref refcnt;	/* Reference counter for the structure */
152 	unsigned int capabilities; /* Device capabilities */
153 	unsigned int min_ratio;
154 	unsigned int max_ratio, max_prop_frac;
155 
156 	/*
157 	 * Sum of avg_write_bw of wbs with dirty inodes.  > 0 if there are
158 	 * any dirty wbs, which is depended upon by bdi_has_dirty().
159 	 */
160 	atomic_long_t tot_write_bandwidth;
161 
162 	struct bdi_writeback wb;  /* the root writeback info for this bdi */
163 	struct list_head wb_list; /* list of all wbs */
164 #ifdef CONFIG_CGROUP_WRITEBACK
165 	struct radix_tree_root cgwb_tree; /* radix tree of active cgroup wbs */
166 	struct rb_root cgwb_congested_tree; /* their congested states */
167 #else
168 	struct bdi_writeback_congested *wb_congested;
169 #endif
170 	wait_queue_head_t wb_waitq;
171 
172 	struct device *dev;
173 	struct device *owner;
174 
175 	struct timer_list laptop_mode_wb_timer;
176 
177 #ifdef CONFIG_DEBUG_FS
178 	struct dentry *debug_dir;
179 	struct dentry *debug_stats;
180 #endif
181 };
182 
183 enum {
184 	BLK_RW_ASYNC	= 0,
185 	BLK_RW_SYNC	= 1,
186 };
187 
188 void clear_wb_congested(struct bdi_writeback_congested *congested, int sync);
189 void set_wb_congested(struct bdi_writeback_congested *congested, int sync);
190 
191 static inline void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
192 {
193 	clear_wb_congested(bdi->wb.congested, sync);
194 }
195 
196 static inline void set_bdi_congested(struct backing_dev_info *bdi, int sync)
197 {
198 	set_wb_congested(bdi->wb.congested, sync);
199 }
200 
201 #ifdef CONFIG_CGROUP_WRITEBACK
202 
203 /**
204  * wb_tryget - try to increment a wb's refcount
205  * @wb: bdi_writeback to get
206  */
207 static inline bool wb_tryget(struct bdi_writeback *wb)
208 {
209 	if (wb != &wb->bdi->wb)
210 		return percpu_ref_tryget(&wb->refcnt);
211 	return true;
212 }
213 
214 /**
215  * wb_get - increment a wb's refcount
216  * @wb: bdi_writeback to get
217  */
218 static inline void wb_get(struct bdi_writeback *wb)
219 {
220 	if (wb != &wb->bdi->wb)
221 		percpu_ref_get(&wb->refcnt);
222 }
223 
224 /**
225  * wb_put - decrement a wb's refcount
226  * @wb: bdi_writeback to put
227  */
228 static inline void wb_put(struct bdi_writeback *wb)
229 {
230 	if (wb != &wb->bdi->wb)
231 		percpu_ref_put(&wb->refcnt);
232 }
233 
234 /**
235  * wb_dying - is a wb dying?
236  * @wb: bdi_writeback of interest
237  *
238  * Returns whether @wb is unlinked and being drained.
239  */
240 static inline bool wb_dying(struct bdi_writeback *wb)
241 {
242 	return percpu_ref_is_dying(&wb->refcnt);
243 }
244 
245 #else	/* CONFIG_CGROUP_WRITEBACK */
246 
247 static inline bool wb_tryget(struct bdi_writeback *wb)
248 {
249 	return true;
250 }
251 
252 static inline void wb_get(struct bdi_writeback *wb)
253 {
254 }
255 
256 static inline void wb_put(struct bdi_writeback *wb)
257 {
258 }
259 
260 static inline bool wb_dying(struct bdi_writeback *wb)
261 {
262 	return false;
263 }
264 
265 #endif	/* CONFIG_CGROUP_WRITEBACK */
266 
267 #endif	/* __LINUX_BACKING_DEV_DEFS_H */
268