Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
66114cad TH |
2 | #ifndef __LINUX_BACKING_DEV_DEFS_H |
3 | #define __LINUX_BACKING_DEV_DEFS_H | |
4 | ||
5 | #include <linux/list.h> | |
52ebea74 TH |
6 | #include <linux/radix-tree.h> |
7 | #include <linux/rbtree.h> | |
66114cad TH |
8 | #include <linux/spinlock.h> |
9 | #include <linux/percpu_counter.h> | |
52ebea74 | 10 | #include <linux/percpu-refcount.h> |
66114cad TH |
11 | #include <linux/flex_proportions.h> |
12 | #include <linux/timer.h> | |
13 | #include <linux/workqueue.h> | |
d03f6cdc | 14 | #include <linux/kref.h> |
66114cad TH |
15 | |
16 | struct page; | |
17 | struct device; | |
18 | struct dentry; | |
19 | ||
20 | /* | |
21 | * Bits in bdi_writeback.state | |
22 | */ | |
23 | enum wb_state { | |
66114cad | 24 | WB_registered, /* bdi_register() was done */ |
5318ce7d | 25 | WB_shutting_down, /* wb_shutdown() in progress */ |
66114cad | 26 | WB_writeback_running, /* Writeback is in progress */ |
d6c10f1f | 27 | WB_has_dirty_io, /* Dirty inodes on ->b_{dirty|io|more_io} */ |
aac8d41c | 28 | WB_start_all, /* nr_pages == 0 (all) work pending */ |
66114cad TH |
29 | }; |
30 | ||
4aa9c692 TH |
31 | enum wb_congested_state { |
32 | WB_async_congested, /* The async (write) queue is getting full */ | |
33 | WB_sync_congested, /* The sync queue is getting full */ | |
34 | }; | |
35 | ||
66114cad TH |
36 | typedef int (congested_fn)(void *, int); |
37 | ||
38 | enum wb_stat_item { | |
39 | WB_RECLAIMABLE, | |
40 | WB_WRITEBACK, | |
41 | WB_DIRTIED, | |
42 | WB_WRITTEN, | |
43 | NR_WB_STAT_ITEMS | |
44 | }; | |
45 | ||
46 | #define WB_STAT_BATCH (8*(1+ilog2(nr_cpu_ids))) | |
47 | ||
85009b4f JA |
48 | /* |
49 | * why some writeback work was initiated | |
50 | */ | |
51 | enum wb_reason { | |
52 | WB_REASON_BACKGROUND, | |
53 | WB_REASON_VMSCAN, | |
54 | WB_REASON_SYNC, | |
55 | WB_REASON_PERIODIC, | |
56 | WB_REASON_LAPTOP_TIMER, | |
57 | WB_REASON_FREE_MORE_MEM, | |
58 | WB_REASON_FS_FREE_SPACE, | |
59 | /* | |
60 | * There is no bdi forker thread any more and works are done | |
61 | * by emergency worker, however, this is TPs userland visible | |
62 | * and we'll be exposing exactly the same information, | |
63 | * so it has a mismatch name. | |
64 | */ | |
65 | WB_REASON_FORKER_THREAD, | |
66 | ||
67 | WB_REASON_MAX, | |
68 | }; | |
69 | ||
52ebea74 TH |
70 | /* |
71 | * For cgroup writeback, multiple wb's may map to the same blkcg. Those | |
72 | * wb's can operate mostly independently but should share the congested | |
73 | * state. To facilitate such sharing, the congested state is tracked using | |
74 | * the following struct which is created on demand, indexed by blkcg ID on | |
75 | * its bdi, and refcounted. | |
76 | */ | |
4aa9c692 TH |
77 | struct bdi_writeback_congested { |
78 | unsigned long state; /* WB_[a]sync_congested flags */ | |
a13f35e8 | 79 | atomic_t refcnt; /* nr of attached wb's and blkg */ |
52ebea74 TH |
80 | |
81 | #ifdef CONFIG_CGROUP_WRITEBACK | |
b7d680d7 JK |
82 | struct backing_dev_info *__bdi; /* the associated bdi, set to NULL |
83 | * on bdi unregistration. For memcg-wb | |
84 | * internal use only! */ | |
52ebea74 TH |
85 | int blkcg_id; /* ID of the associated blkcg */ |
86 | struct rb_node rb_node; /* on bdi->cgwb_congestion_tree */ | |
87 | #endif | |
4aa9c692 TH |
88 | }; |
89 | ||
52ebea74 TH |
90 | /* |
91 | * Each wb (bdi_writeback) can perform writeback operations, is measured | |
92 | * and throttled, independently. Without cgroup writeback, each bdi | |
93 | * (bdi_writeback) is served by its embedded bdi->wb. | |
94 | * | |
95 | * On the default hierarchy, blkcg implicitly enables memcg. This allows | |
96 | * using memcg's page ownership for attributing writeback IOs, and every | |
97 | * memcg - blkcg combination can be served by its own wb by assigning a | |
98 | * dedicated wb to each memcg, which enables isolation across different | |
99 | * cgroups and propagation of IO back pressure down from the IO layer upto | |
100 | * the tasks which are generating the dirty pages to be written back. | |
101 | * | |
102 | * A cgroup wb is indexed on its bdi by the ID of the associated memcg, | |
103 | * refcounted with the number of inodes attached to it, and pins the memcg | |
104 | * and the corresponding blkcg. As the corresponding blkcg for a memcg may | |
105 | * change as blkcg is disabled and enabled higher up in the hierarchy, a wb | |
106 | * is tested for blkcg after lookup and removed from index on mismatch so | |
107 | * that a new wb for the combination can be created. | |
108 | */ | |
66114cad TH |
109 | struct bdi_writeback { |
110 | struct backing_dev_info *bdi; /* our parent bdi */ | |
111 | ||
112 | unsigned long state; /* Always use atomic bitops on this */ | |
113 | unsigned long last_old_flush; /* last old data flush */ | |
114 | ||
115 | struct list_head b_dirty; /* dirty inodes */ | |
116 | struct list_head b_io; /* parked for writeback */ | |
117 | struct list_head b_more_io; /* parked for more writeback */ | |
118 | struct list_head b_dirty_time; /* time stamps are dirty */ | |
119 | spinlock_t list_lock; /* protects the b_* lists */ | |
120 | ||
121 | struct percpu_counter stat[NR_WB_STAT_ITEMS]; | |
122 | ||
4aa9c692 TH |
123 | struct bdi_writeback_congested *congested; |
124 | ||
66114cad TH |
125 | unsigned long bw_time_stamp; /* last time write bw is updated */ |
126 | unsigned long dirtied_stamp; | |
127 | unsigned long written_stamp; /* pages written at bw_time_stamp */ | |
128 | unsigned long write_bandwidth; /* the estimated write bandwidth */ | |
95a46c65 | 129 | unsigned long avg_write_bandwidth; /* further smoothed write bw, > 0 */ |
66114cad TH |
130 | |
131 | /* | |
132 | * The base dirty throttle rate, re-calculated on every 200ms. | |
133 | * All the bdi tasks' dirty rate will be curbed under it. | |
134 | * @dirty_ratelimit tracks the estimated @balanced_dirty_ratelimit | |
135 | * in small steps and is much more smooth/stable than the latter. | |
136 | */ | |
137 | unsigned long dirty_ratelimit; | |
138 | unsigned long balanced_dirty_ratelimit; | |
139 | ||
140 | struct fprop_local_percpu completions; | |
141 | int dirty_exceeded; | |
85009b4f | 142 | enum wb_reason start_all_reason; |
66114cad TH |
143 | |
144 | spinlock_t work_lock; /* protects work_list & dwork scheduling */ | |
145 | struct list_head work_list; | |
146 | struct delayed_work dwork; /* work item used for writeback */ | |
52ebea74 | 147 | |
b57d74af JA |
148 | unsigned long dirty_sleep; /* last wait */ |
149 | ||
b817525a TH |
150 | struct list_head bdi_node; /* anchored at bdi->wb_list */ |
151 | ||
52ebea74 TH |
152 | #ifdef CONFIG_CGROUP_WRITEBACK |
153 | struct percpu_ref refcnt; /* used only for !root wb's */ | |
841710aa | 154 | struct fprop_local_percpu memcg_completions; |
52ebea74 TH |
155 | struct cgroup_subsys_state *memcg_css; /* the associated memcg */ |
156 | struct cgroup_subsys_state *blkcg_css; /* and blkcg */ | |
157 | struct list_head memcg_node; /* anchored at memcg->cgwb_list */ | |
158 | struct list_head blkcg_node; /* anchored at blkcg->cgwb_list */ | |
159 | ||
160 | union { | |
161 | struct work_struct release_work; | |
162 | struct rcu_head rcu; | |
163 | }; | |
164 | #endif | |
66114cad TH |
165 | }; |
166 | ||
167 | struct backing_dev_info { | |
168 | struct list_head bdi_list; | |
ea1754a0 | 169 | unsigned long ra_pages; /* max readahead in PAGE_SIZE units */ |
9491ae4a | 170 | unsigned long io_pages; /* max allowed IO size */ |
66114cad TH |
171 | congested_fn *congested_fn; /* Function pointer if device is md/dm */ |
172 | void *congested_data; /* Pointer to aux data for congested func */ | |
173 | ||
fca39346 | 174 | const char *name; |
66114cad | 175 | |
d03f6cdc | 176 | struct kref refcnt; /* Reference counter for the structure */ |
8db378a5 | 177 | unsigned int capabilities; /* Device capabilities */ |
66114cad TH |
178 | unsigned int min_ratio; |
179 | unsigned int max_ratio, max_prop_frac; | |
180 | ||
95a46c65 TH |
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; | |
766a9d6e | 186 | |
52ebea74 | 187 | struct bdi_writeback wb; /* the root writeback info for this bdi */ |
b817525a | 188 | struct list_head wb_list; /* list of all wbs */ |
52ebea74 TH |
189 | #ifdef CONFIG_CGROUP_WRITEBACK |
190 | struct radix_tree_root cgwb_tree; /* radix tree of active cgroup wbs */ | |
191 | struct rb_root cgwb_congested_tree; /* their congested states */ | |
a13f35e8 TH |
192 | #else |
193 | struct bdi_writeback_congested *wb_congested; | |
52ebea74 | 194 | #endif |
cc395d7f TH |
195 | wait_queue_head_t wb_waitq; |
196 | ||
66114cad | 197 | struct device *dev; |
df08c32c | 198 | struct device *owner; |
66114cad TH |
199 | |
200 | struct timer_list laptop_mode_wb_timer; | |
201 | ||
202 | #ifdef CONFIG_DEBUG_FS | |
203 | struct dentry *debug_dir; | |
204 | struct dentry *debug_stats; | |
205 | #endif | |
206 | }; | |
207 | ||
208 | enum { | |
209 | BLK_RW_ASYNC = 0, | |
210 | BLK_RW_SYNC = 1, | |
211 | }; | |
212 | ||
ec8a6f26 TH |
213 | void clear_wb_congested(struct bdi_writeback_congested *congested, int sync); |
214 | void set_wb_congested(struct bdi_writeback_congested *congested, int sync); | |
215 | ||
216 | static inline void clear_bdi_congested(struct backing_dev_info *bdi, int sync) | |
217 | { | |
218 | clear_wb_congested(bdi->wb.congested, sync); | |
219 | } | |
220 | ||
221 | static inline void set_bdi_congested(struct backing_dev_info *bdi, int sync) | |
222 | { | |
223 | set_wb_congested(bdi->wb.congested, sync); | |
224 | } | |
66114cad | 225 | |
21c6321f TH |
226 | #ifdef CONFIG_CGROUP_WRITEBACK |
227 | ||
228 | /** | |
229 | * wb_tryget - try to increment a wb's refcount | |
230 | * @wb: bdi_writeback to get | |
231 | */ | |
232 | static inline bool wb_tryget(struct bdi_writeback *wb) | |
233 | { | |
234 | if (wb != &wb->bdi->wb) | |
235 | return percpu_ref_tryget(&wb->refcnt); | |
236 | return true; | |
237 | } | |
238 | ||
239 | /** | |
240 | * wb_get - increment a wb's refcount | |
241 | * @wb: bdi_writeback to get | |
242 | */ | |
243 | static inline void wb_get(struct bdi_writeback *wb) | |
244 | { | |
245 | if (wb != &wb->bdi->wb) | |
246 | percpu_ref_get(&wb->refcnt); | |
247 | } | |
248 | ||
249 | /** | |
250 | * wb_put - decrement a wb's refcount | |
251 | * @wb: bdi_writeback to put | |
252 | */ | |
253 | static inline void wb_put(struct bdi_writeback *wb) | |
254 | { | |
255 | if (wb != &wb->bdi->wb) | |
256 | percpu_ref_put(&wb->refcnt); | |
257 | } | |
258 | ||
e8a7abf5 TH |
259 | /** |
260 | * wb_dying - is a wb dying? | |
261 | * @wb: bdi_writeback of interest | |
262 | * | |
263 | * Returns whether @wb is unlinked and being drained. | |
264 | */ | |
265 | static inline bool wb_dying(struct bdi_writeback *wb) | |
266 | { | |
267 | return percpu_ref_is_dying(&wb->refcnt); | |
268 | } | |
269 | ||
21c6321f TH |
270 | #else /* CONFIG_CGROUP_WRITEBACK */ |
271 | ||
272 | static inline bool wb_tryget(struct bdi_writeback *wb) | |
273 | { | |
274 | return true; | |
275 | } | |
276 | ||
277 | static inline void wb_get(struct bdi_writeback *wb) | |
278 | { | |
279 | } | |
280 | ||
281 | static inline void wb_put(struct bdi_writeback *wb) | |
282 | { | |
283 | } | |
284 | ||
e8a7abf5 TH |
285 | static inline bool wb_dying(struct bdi_writeback *wb) |
286 | { | |
287 | return false; | |
288 | } | |
289 | ||
21c6321f TH |
290 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
291 | ||
66114cad | 292 | #endif /* __LINUX_BACKING_DEV_DEFS_H */ |