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