• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 	struct list_head bdi_node;	/* anchored at bdi->wb_list */
120 
121 #ifdef CONFIG_CGROUP_WRITEBACK
122 	struct percpu_ref refcnt;	/* used only for !root wb's */
123 	struct fprop_local_percpu memcg_completions;
124 	struct cgroup_subsys_state *memcg_css; /* the associated memcg */
125 	struct cgroup_subsys_state *blkcg_css; /* and blkcg */
126 	struct list_head memcg_node;	/* anchored at memcg->cgwb_list */
127 	struct list_head blkcg_node;	/* anchored at blkcg->cgwb_list */
128 
129 	union {
130 		struct work_struct release_work;
131 		struct rcu_head rcu;
132 	};
133 #endif
134 };
135 
136 struct backing_dev_info {
137 	struct list_head bdi_list;
138 	unsigned long ra_pages;	/* max readahead in PAGE_SIZE units */
139 	unsigned int capabilities; /* Device capabilities */
140 	congested_fn *congested_fn; /* Function pointer if device is md/dm */
141 	void *congested_data;	/* Pointer to aux data for congested func */
142 
143 	char *name;
144 
145 	unsigned int min_ratio;
146 	unsigned int max_ratio, max_prop_frac;
147 
148 	/*
149 	 * Sum of avg_write_bw of wbs with dirty inodes.  > 0 if there are
150 	 * any dirty wbs, which is depended upon by bdi_has_dirty().
151 	 */
152 	atomic_long_t tot_write_bandwidth;
153 
154 	struct bdi_writeback wb;  /* the root writeback info for this bdi */
155 	struct list_head wb_list; /* list of all wbs */
156 #ifdef CONFIG_CGROUP_WRITEBACK
157 	struct radix_tree_root cgwb_tree; /* radix tree of active cgroup wbs */
158 	struct rb_root cgwb_congested_tree; /* their congested states */
159 	atomic_t usage_cnt; /* counts both cgwbs and cgwb_contested's */
160 #else
161 	struct bdi_writeback_congested *wb_congested;
162 #endif
163 	wait_queue_head_t wb_waitq;
164 
165 	struct device *dev;
166 	struct device *owner;
167 
168 	struct timer_list laptop_mode_wb_timer;
169 
170 #ifdef CONFIG_DEBUG_FS
171 	struct dentry *debug_dir;
172 	struct dentry *debug_stats;
173 #endif
174 };
175 
176 enum {
177 	BLK_RW_ASYNC	= 0,
178 	BLK_RW_SYNC	= 1,
179 };
180 
181 void clear_wb_congested(struct bdi_writeback_congested *congested, int sync);
182 void set_wb_congested(struct bdi_writeback_congested *congested, int sync);
183 
clear_bdi_congested(struct backing_dev_info * bdi,int sync)184 static inline void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
185 {
186 	clear_wb_congested(bdi->wb.congested, sync);
187 }
188 
set_bdi_congested(struct backing_dev_info * bdi,int sync)189 static inline void set_bdi_congested(struct backing_dev_info *bdi, int sync)
190 {
191 	set_wb_congested(bdi->wb.congested, sync);
192 }
193 
194 struct wb_lock_cookie {
195 	bool locked;
196 	unsigned long flags;
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  */
wb_tryget(struct bdi_writeback * wb)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  */
wb_get(struct bdi_writeback * wb)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  */
wb_put(struct bdi_writeback * wb)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  */
wb_dying(struct bdi_writeback * wb)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 
wb_tryget(struct bdi_writeback * wb)245 static inline bool wb_tryget(struct bdi_writeback *wb)
246 {
247 	return true;
248 }
249 
wb_get(struct bdi_writeback * wb)250 static inline void wb_get(struct bdi_writeback *wb)
251 {
252 }
253 
wb_put(struct bdi_writeback * wb)254 static inline void wb_put(struct bdi_writeback *wb)
255 {
256 }
257 
wb_dying(struct bdi_writeback * wb)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 */
266