]>
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} */ |
66114cad TH |
28 | }; |
29 | ||
4aa9c692 TH |
30 | enum wb_congested_state { |
31 | WB_async_congested, /* The async (write) queue is getting full */ | |
32 | WB_sync_congested, /* The sync queue is getting full */ | |
33 | }; | |
34 | ||
66114cad TH |
35 | typedef int (congested_fn)(void *, int); |
36 | ||
37 | enum wb_stat_item { | |
38 | WB_RECLAIMABLE, | |
39 | WB_WRITEBACK, | |
40 | WB_DIRTIED, | |
41 | WB_WRITTEN, | |
42 | NR_WB_STAT_ITEMS | |
43 | }; | |
44 | ||
45 | #define WB_STAT_BATCH (8*(1+ilog2(nr_cpu_ids))) | |
46 | ||
52ebea74 TH |
47 | /* |
48 | * For cgroup writeback, multiple wb's may map to the same blkcg. Those | |
49 | * wb's can operate mostly independently but should share the congested | |
50 | * state. To facilitate such sharing, the congested state is tracked using | |
51 | * the following struct which is created on demand, indexed by blkcg ID on | |
52 | * its bdi, and refcounted. | |
53 | */ | |
4aa9c692 TH |
54 | struct bdi_writeback_congested { |
55 | unsigned long state; /* WB_[a]sync_congested flags */ | |
a13f35e8 | 56 | atomic_t refcnt; /* nr of attached wb's and blkg */ |
52ebea74 TH |
57 | |
58 | #ifdef CONFIG_CGROUP_WRITEBACK | |
b7d680d7 JK |
59 | struct backing_dev_info *__bdi; /* the associated bdi, set to NULL |
60 | * on bdi unregistration. For memcg-wb | |
61 | * internal use only! */ | |
52ebea74 TH |
62 | int blkcg_id; /* ID of the associated blkcg */ |
63 | struct rb_node rb_node; /* on bdi->cgwb_congestion_tree */ | |
64 | #endif | |
4aa9c692 TH |
65 | }; |
66 | ||
52ebea74 TH |
67 | /* |
68 | * Each wb (bdi_writeback) can perform writeback operations, is measured | |
69 | * and throttled, independently. Without cgroup writeback, each bdi | |
70 | * (bdi_writeback) is served by its embedded bdi->wb. | |
71 | * | |
72 | * On the default hierarchy, blkcg implicitly enables memcg. This allows | |
73 | * using memcg's page ownership for attributing writeback IOs, and every | |
74 | * memcg - blkcg combination can be served by its own wb by assigning a | |
75 | * dedicated wb to each memcg, which enables isolation across different | |
76 | * cgroups and propagation of IO back pressure down from the IO layer upto | |
77 | * the tasks which are generating the dirty pages to be written back. | |
78 | * | |
79 | * A cgroup wb is indexed on its bdi by the ID of the associated memcg, | |
80 | * refcounted with the number of inodes attached to it, and pins the memcg | |
81 | * and the corresponding blkcg. As the corresponding blkcg for a memcg may | |
82 | * change as blkcg is disabled and enabled higher up in the hierarchy, a wb | |
83 | * is tested for blkcg after lookup and removed from index on mismatch so | |
84 | * that a new wb for the combination can be created. | |
85 | */ | |
66114cad TH |
86 | struct bdi_writeback { |
87 | struct backing_dev_info *bdi; /* our parent bdi */ | |
88 | ||
89 | unsigned long state; /* Always use atomic bitops on this */ | |
90 | unsigned long last_old_flush; /* last old data flush */ | |
91 | ||
92 | struct list_head b_dirty; /* dirty inodes */ | |
93 | struct list_head b_io; /* parked for writeback */ | |
94 | struct list_head b_more_io; /* parked for more writeback */ | |
95 | struct list_head b_dirty_time; /* time stamps are dirty */ | |
96 | spinlock_t list_lock; /* protects the b_* lists */ | |
97 | ||
98 | struct percpu_counter stat[NR_WB_STAT_ITEMS]; | |
99 | ||
4aa9c692 TH |
100 | struct bdi_writeback_congested *congested; |
101 | ||
66114cad TH |
102 | unsigned long bw_time_stamp; /* last time write bw is updated */ |
103 | unsigned long dirtied_stamp; | |
104 | unsigned long written_stamp; /* pages written at bw_time_stamp */ | |
105 | unsigned long write_bandwidth; /* the estimated write bandwidth */ | |
95a46c65 | 106 | unsigned long avg_write_bandwidth; /* further smoothed write bw, > 0 */ |
66114cad TH |
107 | |
108 | /* | |
109 | * The base dirty throttle rate, re-calculated on every 200ms. | |
110 | * All the bdi tasks' dirty rate will be curbed under it. | |
111 | * @dirty_ratelimit tracks the estimated @balanced_dirty_ratelimit | |
112 | * in small steps and is much more smooth/stable than the latter. | |
113 | */ | |
114 | unsigned long dirty_ratelimit; | |
115 | unsigned long balanced_dirty_ratelimit; | |
116 | ||
117 | struct fprop_local_percpu completions; | |
118 | int dirty_exceeded; | |
119 | ||
120 | spinlock_t work_lock; /* protects work_list & dwork scheduling */ | |
121 | struct list_head work_list; | |
122 | struct delayed_work dwork; /* work item used for writeback */ | |
52ebea74 | 123 | |
b57d74af JA |
124 | unsigned long dirty_sleep; /* last wait */ |
125 | ||
b817525a TH |
126 | struct list_head bdi_node; /* anchored at bdi->wb_list */ |
127 | ||
52ebea74 TH |
128 | #ifdef CONFIG_CGROUP_WRITEBACK |
129 | struct percpu_ref refcnt; /* used only for !root wb's */ | |
841710aa | 130 | struct fprop_local_percpu memcg_completions; |
52ebea74 TH |
131 | struct cgroup_subsys_state *memcg_css; /* the associated memcg */ |
132 | struct cgroup_subsys_state *blkcg_css; /* and blkcg */ | |
133 | struct list_head memcg_node; /* anchored at memcg->cgwb_list */ | |
134 | struct list_head blkcg_node; /* anchored at blkcg->cgwb_list */ | |
135 | ||
136 | union { | |
137 | struct work_struct release_work; | |
138 | struct rcu_head rcu; | |
139 | }; | |
140 | #endif | |
66114cad TH |
141 | }; |
142 | ||
143 | struct backing_dev_info { | |
144 | struct list_head bdi_list; | |
ea1754a0 | 145 | unsigned long ra_pages; /* max readahead in PAGE_SIZE units */ |
9491ae4a | 146 | unsigned long io_pages; /* max allowed IO size */ |
66114cad TH |
147 | congested_fn *congested_fn; /* Function pointer if device is md/dm */ |
148 | void *congested_data; /* Pointer to aux data for congested func */ | |
149 | ||
fca39346 | 150 | const char *name; |
66114cad | 151 | |
d03f6cdc | 152 | struct kref refcnt; /* Reference counter for the structure */ |
8db378a5 | 153 | unsigned int capabilities; /* Device capabilities */ |
66114cad TH |
154 | unsigned int min_ratio; |
155 | unsigned int max_ratio, max_prop_frac; | |
156 | ||
95a46c65 TH |
157 | /* |
158 | * Sum of avg_write_bw of wbs with dirty inodes. > 0 if there are | |
159 | * any dirty wbs, which is depended upon by bdi_has_dirty(). | |
160 | */ | |
161 | atomic_long_t tot_write_bandwidth; | |
766a9d6e | 162 | |
52ebea74 | 163 | struct bdi_writeback wb; /* the root writeback info for this bdi */ |
b817525a | 164 | struct list_head wb_list; /* list of all wbs */ |
52ebea74 TH |
165 | #ifdef CONFIG_CGROUP_WRITEBACK |
166 | struct radix_tree_root cgwb_tree; /* radix tree of active cgroup wbs */ | |
167 | struct rb_root cgwb_congested_tree; /* their congested states */ | |
a13f35e8 TH |
168 | #else |
169 | struct bdi_writeback_congested *wb_congested; | |
52ebea74 | 170 | #endif |
cc395d7f TH |
171 | wait_queue_head_t wb_waitq; |
172 | ||
66114cad | 173 | struct device *dev; |
df08c32c | 174 | struct device *owner; |
66114cad TH |
175 | |
176 | struct timer_list laptop_mode_wb_timer; | |
177 | ||
178 | #ifdef CONFIG_DEBUG_FS | |
179 | struct dentry *debug_dir; | |
180 | struct dentry *debug_stats; | |
181 | #endif | |
182 | }; | |
183 | ||
184 | enum { | |
185 | BLK_RW_ASYNC = 0, | |
186 | BLK_RW_SYNC = 1, | |
187 | }; | |
188 | ||
ec8a6f26 TH |
189 | void clear_wb_congested(struct bdi_writeback_congested *congested, int sync); |
190 | void set_wb_congested(struct bdi_writeback_congested *congested, int sync); | |
191 | ||
192 | static inline void clear_bdi_congested(struct backing_dev_info *bdi, int sync) | |
193 | { | |
194 | clear_wb_congested(bdi->wb.congested, sync); | |
195 | } | |
196 | ||
197 | static inline void set_bdi_congested(struct backing_dev_info *bdi, int sync) | |
198 | { | |
199 | set_wb_congested(bdi->wb.congested, sync); | |
200 | } | |
66114cad | 201 | |
21c6321f TH |
202 | #ifdef CONFIG_CGROUP_WRITEBACK |
203 | ||
204 | /** | |
205 | * wb_tryget - try to increment a wb's refcount | |
206 | * @wb: bdi_writeback to get | |
207 | */ | |
208 | static inline bool wb_tryget(struct bdi_writeback *wb) | |
209 | { | |
210 | if (wb != &wb->bdi->wb) | |
211 | return percpu_ref_tryget(&wb->refcnt); | |
212 | return true; | |
213 | } | |
214 | ||
215 | /** | |
216 | * wb_get - increment a wb's refcount | |
217 | * @wb: bdi_writeback to get | |
218 | */ | |
219 | static inline void wb_get(struct bdi_writeback *wb) | |
220 | { | |
221 | if (wb != &wb->bdi->wb) | |
222 | percpu_ref_get(&wb->refcnt); | |
223 | } | |
224 | ||
225 | /** | |
226 | * wb_put - decrement a wb's refcount | |
227 | * @wb: bdi_writeback to put | |
228 | */ | |
229 | static inline void wb_put(struct bdi_writeback *wb) | |
230 | { | |
231 | if (wb != &wb->bdi->wb) | |
232 | percpu_ref_put(&wb->refcnt); | |
233 | } | |
234 | ||
e8a7abf5 TH |
235 | /** |
236 | * wb_dying - is a wb dying? | |
237 | * @wb: bdi_writeback of interest | |
238 | * | |
239 | * Returns whether @wb is unlinked and being drained. | |
240 | */ | |
241 | static inline bool wb_dying(struct bdi_writeback *wb) | |
242 | { | |
243 | return percpu_ref_is_dying(&wb->refcnt); | |
244 | } | |
245 | ||
21c6321f TH |
246 | #else /* CONFIG_CGROUP_WRITEBACK */ |
247 | ||
248 | static inline bool wb_tryget(struct bdi_writeback *wb) | |
249 | { | |
250 | return true; | |
251 | } | |
252 | ||
253 | static inline void wb_get(struct bdi_writeback *wb) | |
254 | { | |
255 | } | |
256 | ||
257 | static inline void wb_put(struct bdi_writeback *wb) | |
258 | { | |
259 | } | |
260 | ||
e8a7abf5 TH |
261 | static inline bool wb_dying(struct bdi_writeback *wb) |
262 | { | |
263 | return false; | |
264 | } | |
265 | ||
21c6321f TH |
266 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
267 | ||
66114cad | 268 | #endif /* __LINUX_BACKING_DEV_DEFS_H */ |