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
17 struct page;
18 struct device;
19 struct dentry;
20
21 /*
22 * Bits in bdi_writeback.state
23 */
24 enum wb_state {
25 WB_registered, /* bdi_register() was done */
26 WB_writeback_running, /* Writeback is in progress */
27 WB_has_dirty_io, /* Dirty inodes on ->b_{dirty|io|more_io} */
28 WB_start_all, /* nr_pages == 0 (all) work pending */
29 };
30
31 enum wb_stat_item {
32 WB_RECLAIMABLE,
33 WB_WRITEBACK,
34 WB_DIRTIED,
35 WB_WRITTEN,
36 NR_WB_STAT_ITEMS
37 };
38
39 #define WB_STAT_BATCH (8*(1+ilog2(nr_cpu_ids)))
40
41 /*
42 * why some writeback work was initiated
43 */
44 enum wb_reason {
45 WB_REASON_BACKGROUND,
46 WB_REASON_VMSCAN,
47 WB_REASON_SYNC,
48 WB_REASON_PERIODIC,
49 WB_REASON_FS_FREE_SPACE,
50 /*
51 * There is no bdi forker thread any more and works are done
52 * by emergency worker, however, this is TPs userland visible
53 * and we'll be exposing exactly the same information,
54 * so it has a mismatch name.
55 */
56 WB_REASON_FORKER_THREAD,
57 WB_REASON_FOREIGN_FLUSH,
58
59 WB_REASON_MAX,
60 };
61
62 struct wb_completion {
63 atomic_t cnt;
64 wait_queue_head_t *waitq;
65 unsigned long progress_stamp; /* The jiffies when slow progress is detected */
66 unsigned long wait_start; /* The jiffies when waiting for the writeback work to finish */
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 struct list_head bdi_node; /* anchored at bdi->wb_list */
146
147 #ifdef CONFIG_CGROUP_WRITEBACK
148 struct percpu_ref refcnt; /* used only for !root wb's */
149 struct fprop_local_percpu memcg_completions;
150 struct cgroup_subsys_state *memcg_css; /* the associated memcg */
151 struct cgroup_subsys_state *blkcg_css; /* and blkcg */
152 struct list_head memcg_node; /* anchored at memcg->cgwb_list */
153 struct list_head blkcg_node; /* anchored at blkcg->cgwb_list */
154 struct list_head b_attached; /* attached inodes, protected by list_lock */
155 struct list_head offline_node; /* anchored at offline_cgwbs */
156 struct work_struct switch_work; /* work used to perform inode switching
157 * to this wb */
158 struct llist_head switch_wbs_ctxs; /* queued contexts for
159 * writeback switching */
160
161 union {
162 struct work_struct release_work;
163 struct rcu_head rcu;
164 };
165 #endif
166 };
167
168 struct backing_dev_info {
169 u64 id;
170 struct rb_node rb_node; /* keyed by ->id */
171 struct list_head bdi_list;
172 /* max readahead in PAGE_SIZE units */
173 unsigned long __data_racy ra_pages;
174
175 unsigned long io_pages; /* max allowed IO size */
176
177 struct kref refcnt; /* Reference counter for the structure */
178 unsigned int capabilities; /* Device capabilities */
179 unsigned int min_ratio;
180 unsigned int max_ratio, max_prop_frac;
181
182 /*
183 * Sum of avg_write_bw of wbs with dirty inodes. > 0 if there are
184 * any dirty wbs, which is depended upon by bdi_has_dirty().
185 */
186 atomic_long_t tot_write_bandwidth;
187 /*
188 * Jiffies when last process was dirty throttled on this bdi. Used by
189 * blk-wbt.
190 */
191 unsigned long last_bdp_sleep;
192
193 struct bdi_writeback wb; /* the root writeback info for this bdi */
194 struct list_head wb_list; /* list of all wbs */
195 #ifdef CONFIG_CGROUP_WRITEBACK
196 struct radix_tree_root cgwb_tree; /* radix tree of active cgroup wbs */
197 struct mutex cgwb_release_mutex; /* protect shutdown of wb structs */
198 struct rw_semaphore wb_switch_rwsem; /* no cgwb switch while syncing */
199 #endif
200 wait_queue_head_t wb_waitq;
201
202 struct device *dev;
203 char dev_name[64];
204 struct device *owner;
205
206 #ifdef CONFIG_DEBUG_FS
207 struct dentry *debug_dir;
208 #endif
209 };
210
211 struct wb_lock_cookie {
212 bool locked;
213 unsigned long flags;
214 };
215
216 #ifdef CONFIG_CGROUP_WRITEBACK
217
218 /**
219 * wb_tryget - try to increment a wb's refcount
220 * @wb: bdi_writeback to get
221 */
wb_tryget(struct bdi_writeback * wb)222 static inline bool wb_tryget(struct bdi_writeback *wb)
223 {
224 if (wb != &wb->bdi->wb)
225 return percpu_ref_tryget(&wb->refcnt);
226 return true;
227 }
228
229 /**
230 * wb_get - increment a wb's refcount
231 * @wb: bdi_writeback to get
232 */
wb_get(struct bdi_writeback * wb)233 static inline void wb_get(struct bdi_writeback *wb)
234 {
235 if (wb != &wb->bdi->wb)
236 percpu_ref_get(&wb->refcnt);
237 }
238
239 /**
240 * wb_put - decrement a wb's refcount
241 * @wb: bdi_writeback to put
242 * @nr: number of references to put
243 */
wb_put_many(struct bdi_writeback * wb,unsigned long nr)244 static inline void wb_put_many(struct bdi_writeback *wb, unsigned long nr)
245 {
246 if (WARN_ON_ONCE(!wb->bdi)) {
247 /*
248 * A driver bug might cause a file to be removed before bdi was
249 * initialized.
250 */
251 return;
252 }
253
254 if (wb != &wb->bdi->wb)
255 percpu_ref_put_many(&wb->refcnt, nr);
256 }
257
258 /**
259 * wb_put - decrement a wb's refcount
260 * @wb: bdi_writeback to put
261 */
wb_put(struct bdi_writeback * wb)262 static inline void wb_put(struct bdi_writeback *wb)
263 {
264 wb_put_many(wb, 1);
265 }
266
267 /**
268 * wb_dying - is a wb dying?
269 * @wb: bdi_writeback of interest
270 *
271 * Returns whether @wb is unlinked and being drained.
272 */
wb_dying(struct bdi_writeback * wb)273 static inline bool wb_dying(struct bdi_writeback *wb)
274 {
275 return percpu_ref_is_dying(&wb->refcnt);
276 }
277
278 #else /* CONFIG_CGROUP_WRITEBACK */
279
wb_tryget(struct bdi_writeback * wb)280 static inline bool wb_tryget(struct bdi_writeback *wb)
281 {
282 return true;
283 }
284
wb_get(struct bdi_writeback * wb)285 static inline void wb_get(struct bdi_writeback *wb)
286 {
287 }
288
wb_put(struct bdi_writeback * wb)289 static inline void wb_put(struct bdi_writeback *wb)
290 {
291 }
292
wb_put_many(struct bdi_writeback * wb,unsigned long nr)293 static inline void wb_put_many(struct bdi_writeback *wb, unsigned long nr)
294 {
295 }
296
wb_dying(struct bdi_writeback * wb)297 static inline bool wb_dying(struct bdi_writeback *wb)
298 {
299 return false;
300 }
301
302 #endif /* CONFIG_CGROUP_WRITEBACK */
303
304 #endif /* __LINUX_BACKING_DEV_DEFS_H */
305