blob: 1b4d69f68c33cc73ad99a1136b2408c71e763fb8 [file] [log] [blame]
Tejun Heo66114ca2015-05-22 17:13:32 -04001#ifndef __LINUX_BACKING_DEV_DEFS_H
2#define __LINUX_BACKING_DEV_DEFS_H
3
4#include <linux/list.h>
Tejun Heo52ebea72015-05-22 17:13:37 -04005#include <linux/radix-tree.h>
6#include <linux/rbtree.h>
Tejun Heo66114ca2015-05-22 17:13:32 -04007#include <linux/spinlock.h>
8#include <linux/percpu_counter.h>
Tejun Heo52ebea72015-05-22 17:13:37 -04009#include <linux/percpu-refcount.h>
Tejun Heo66114ca2015-05-22 17:13:32 -040010#include <linux/flex_proportions.h>
11#include <linux/timer.h>
12#include <linux/workqueue.h>
13
14struct page;
15struct device;
16struct dentry;
17
18/*
19 * Bits in bdi_writeback.state
20 */
21enum wb_state {
Tejun Heo66114ca2015-05-22 17:13:32 -040022 WB_registered, /* bdi_register() was done */
23 WB_writeback_running, /* Writeback is in progress */
Tejun Heod6c10f12015-05-22 17:13:45 -040024 WB_has_dirty_io, /* Dirty inodes on ->b_{dirty|io|more_io} */
Tejun Heo66114ca2015-05-22 17:13:32 -040025};
26
Tejun Heo4aa9c692015-05-22 17:13:35 -040027enum 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
Tejun Heo66114ca2015-05-22 17:13:32 -040032typedef int (congested_fn)(void *, int);
33
34enum 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
Tejun Heo52ebea72015-05-22 17:13:37 -040044/*
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 */
Tejun Heo4aa9c692015-05-22 17:13:35 -040051struct bdi_writeback_congested {
52 unsigned long state; /* WB_[a]sync_congested flags */
Tejun Heoa13f35e2015-07-02 08:44:34 -060053 atomic_t refcnt; /* nr of attached wb's and blkg */
Tejun Heo52ebea72015-05-22 17:13:37 -040054
55#ifdef CONFIG_CGROUP_WRITEBACK
56 struct backing_dev_info *bdi; /* the associated bdi */
Tejun Heo52ebea72015-05-22 17:13:37 -040057 int blkcg_id; /* ID of the associated blkcg */
58 struct rb_node rb_node; /* on bdi->cgwb_congestion_tree */
59#endif
Tejun Heo4aa9c692015-05-22 17:13:35 -040060};
61
Tejun Heo52ebea72015-05-22 17:13:37 -040062/*
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 */
Tejun Heo66114ca2015-05-22 17:13:32 -040081struct 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
Tejun Heo4aa9c692015-05-22 17:13:35 -040095 struct bdi_writeback_congested *congested;
96
Tejun Heo66114ca2015-05-22 17:13:32 -040097 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 */
Tejun Heo95a46c62015-05-22 17:13:47 -0400101 unsigned long avg_write_bandwidth; /* further smoothed write bw, > 0 */
Tejun Heo66114ca2015-05-22 17:13:32 -0400102
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 */
Tejun Heo52ebea72015-05-22 17:13:37 -0400118
Tejun Heob8175252015-10-02 14:47:05 -0400119 struct list_head bdi_node; /* anchored at bdi->wb_list */
120
Tejun Heo52ebea72015-05-22 17:13:37 -0400121#ifdef CONFIG_CGROUP_WRITEBACK
122 struct percpu_ref refcnt; /* used only for !root wb's */
Tejun Heo841710a2015-05-22 18:23:33 -0400123 struct fprop_local_percpu memcg_completions;
Tejun Heo52ebea72015-05-22 17:13:37 -0400124 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
Tejun Heo66114ca2015-05-22 17:13:32 -0400134};
135
136struct backing_dev_info {
137 struct list_head bdi_list;
138 unsigned long ra_pages; /* max readahead in PAGE_CACHE_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
Tejun Heo95a46c62015-05-22 17:13:47 -0400148 /*
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;
Tejun Heo766a9d62015-05-22 17:13:46 -0400153
Tejun Heo52ebea72015-05-22 17:13:37 -0400154 struct bdi_writeback wb; /* the root writeback info for this bdi */
Tejun Heob8175252015-10-02 14:47:05 -0400155 struct list_head wb_list; /* list of all wbs */
Tejun Heo52ebea72015-05-22 17:13:37 -0400156#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 */
Tejun Heoa13f35e2015-07-02 08:44:34 -0600160#else
161 struct bdi_writeback_congested *wb_congested;
Tejun Heo52ebea72015-05-22 17:13:37 -0400162#endif
Tejun Heocc395d72015-05-22 17:13:58 -0400163 wait_queue_head_t wb_waitq;
164
Tejun Heo66114ca2015-05-22 17:13:32 -0400165 struct device *dev;
166
167 struct timer_list laptop_mode_wb_timer;
168
169#ifdef CONFIG_DEBUG_FS
170 struct dentry *debug_dir;
171 struct dentry *debug_stats;
172#endif
173};
174
175enum {
176 BLK_RW_ASYNC = 0,
177 BLK_RW_SYNC = 1,
178};
179
Tejun Heoec8a6f22015-05-22 17:13:41 -0400180void clear_wb_congested(struct bdi_writeback_congested *congested, int sync);
181void set_wb_congested(struct bdi_writeback_congested *congested, int sync);
182
183static inline void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
184{
185 clear_wb_congested(bdi->wb.congested, sync);
186}
187
188static inline void set_bdi_congested(struct backing_dev_info *bdi, int sync)
189{
190 set_wb_congested(bdi->wb.congested, sync);
191}
Tejun Heo66114ca2015-05-22 17:13:32 -0400192
Tejun Heo21c63212015-05-28 14:50:49 -0400193#ifdef CONFIG_CGROUP_WRITEBACK
194
195/**
196 * wb_tryget - try to increment a wb's refcount
197 * @wb: bdi_writeback to get
198 */
199static inline bool wb_tryget(struct bdi_writeback *wb)
200{
201 if (wb != &wb->bdi->wb)
202 return percpu_ref_tryget(&wb->refcnt);
203 return true;
204}
205
206/**
207 * wb_get - increment a wb's refcount
208 * @wb: bdi_writeback to get
209 */
210static inline void wb_get(struct bdi_writeback *wb)
211{
212 if (wb != &wb->bdi->wb)
213 percpu_ref_get(&wb->refcnt);
214}
215
216/**
217 * wb_put - decrement a wb's refcount
218 * @wb: bdi_writeback to put
219 */
220static inline void wb_put(struct bdi_writeback *wb)
221{
222 if (wb != &wb->bdi->wb)
223 percpu_ref_put(&wb->refcnt);
224}
225
Tejun Heoe8a7abf2015-05-28 14:50:57 -0400226/**
227 * wb_dying - is a wb dying?
228 * @wb: bdi_writeback of interest
229 *
230 * Returns whether @wb is unlinked and being drained.
231 */
232static inline bool wb_dying(struct bdi_writeback *wb)
233{
234 return percpu_ref_is_dying(&wb->refcnt);
235}
236
Tejun Heo21c63212015-05-28 14:50:49 -0400237#else /* CONFIG_CGROUP_WRITEBACK */
238
239static inline bool wb_tryget(struct bdi_writeback *wb)
240{
241 return true;
242}
243
244static inline void wb_get(struct bdi_writeback *wb)
245{
246}
247
248static inline void wb_put(struct bdi_writeback *wb)
249{
250}
251
Tejun Heoe8a7abf2015-05-28 14:50:57 -0400252static inline bool wb_dying(struct bdi_writeback *wb)
253{
254 return false;
255}
256
Tejun Heo21c63212015-05-28 14:50:49 -0400257#endif /* CONFIG_CGROUP_WRITEBACK */
258
Tejun Heo66114ca2015-05-22 17:13:32 -0400259#endif /* __LINUX_BACKING_DEV_DEFS_H */