1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __LINUX_BACKING_DEV_DEFS_H
3 #define __LINUX_BACKING_DEV_DEFS_H
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>
22 * Bits in bdi_writeback.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 */
39 #define WB_STAT_BATCH (8*(1+ilog2(nr_cpu_ids)))
42 * why some writeback work was initiated
49 WB_REASON_LAPTOP_TIMER,
50 WB_REASON_FS_FREE_SPACE,
52 * There is no bdi forker thread any more and works are done
53 * by emergency worker, however, this is TPs userland visible
54 * and we'll be exposing exactly the same information,
55 * so it has a mismatch name.
57 WB_REASON_FORKER_THREAD,
58 WB_REASON_FOREIGN_FLUSH,
63 struct wb_completion {
65 wait_queue_head_t *waitq;
68 #define __WB_COMPLETION_INIT(_waitq) \
69 (struct wb_completion){ .cnt = ATOMIC_INIT(1), .waitq = (_waitq) }
72 * If one wants to wait for one or more wb_writeback_works, each work's
73 * ->done should be set to a wb_completion defined using the following
74 * macro. Once all work items are issued with wb_queue_work(), the caller
75 * can wait for the completion of all using wb_wait_for_completion(). Work
76 * items which are waited upon aren't freed automatically on completion.
78 #define WB_COMPLETION_INIT(bdi) __WB_COMPLETION_INIT(&(bdi)->wb_waitq)
80 #define DEFINE_WB_COMPLETION(cmpl, bdi) \
81 struct wb_completion cmpl = WB_COMPLETION_INIT(bdi)
84 * Each wb (bdi_writeback) can perform writeback operations, is measured
85 * and throttled, independently. Without cgroup writeback, each bdi
86 * (bdi_writeback) is served by its embedded bdi->wb.
88 * On the default hierarchy, blkcg implicitly enables memcg. This allows
89 * using memcg's page ownership for attributing writeback IOs, and every
90 * memcg - blkcg combination can be served by its own wb by assigning a
91 * dedicated wb to each memcg, which enables isolation across different
92 * cgroups and propagation of IO back pressure down from the IO layer upto
93 * the tasks which are generating the dirty pages to be written back.
95 * A cgroup wb is indexed on its bdi by the ID of the associated memcg,
96 * refcounted with the number of inodes attached to it, and pins the memcg
97 * and the corresponding blkcg. As the corresponding blkcg for a memcg may
98 * change as blkcg is disabled and enabled higher up in the hierarchy, a wb
99 * is tested for blkcg after lookup and removed from index on mismatch so
100 * that a new wb for the combination can be created.
102 * Each bdi_writeback that is not embedded into the backing_dev_info must hold
103 * a reference to the parent backing_dev_info. See cgwb_create() for details.
105 struct bdi_writeback {
106 struct backing_dev_info *bdi; /* our parent bdi */
108 unsigned long state; /* Always use atomic bitops on this */
109 unsigned long last_old_flush; /* last old data flush */
111 struct list_head b_dirty; /* dirty inodes */
112 struct list_head b_io; /* parked for writeback */
113 struct list_head b_more_io; /* parked for more writeback */
114 struct list_head b_dirty_time; /* time stamps are dirty */
115 spinlock_t list_lock; /* protects the b_* lists */
117 atomic_t writeback_inodes; /* number of inodes under writeback */
118 struct percpu_counter stat[NR_WB_STAT_ITEMS];
120 unsigned long bw_time_stamp; /* last time write bw is updated */
121 unsigned long dirtied_stamp;
122 unsigned long written_stamp; /* pages written at bw_time_stamp */
123 unsigned long write_bandwidth; /* the estimated write bandwidth */
124 unsigned long avg_write_bandwidth; /* further smoothed write bw, > 0 */
127 * The base dirty throttle rate, re-calculated on every 200ms.
128 * All the bdi tasks' dirty rate will be curbed under it.
129 * @dirty_ratelimit tracks the estimated @balanced_dirty_ratelimit
130 * in small steps and is much more smooth/stable than the latter.
132 unsigned long dirty_ratelimit;
133 unsigned long balanced_dirty_ratelimit;
135 struct fprop_local_percpu completions;
137 enum wb_reason start_all_reason;
139 spinlock_t work_lock; /* protects work_list & dwork scheduling */
140 struct list_head work_list;
141 struct delayed_work dwork; /* work item used for writeback */
142 struct delayed_work bw_dwork; /* work item used for bandwidth estimate */
144 unsigned long dirty_sleep; /* last wait */
146 struct list_head bdi_node; /* anchored at bdi->wb_list */
148 #ifdef CONFIG_CGROUP_WRITEBACK
149 struct percpu_ref refcnt; /* used only for !root wb's */
150 struct fprop_local_percpu memcg_completions;
151 struct cgroup_subsys_state *memcg_css; /* the associated memcg */
152 struct cgroup_subsys_state *blkcg_css; /* and blkcg */
153 struct list_head memcg_node; /* anchored at memcg->cgwb_list */
154 struct list_head blkcg_node; /* anchored at blkcg->cgwb_list */
155 struct list_head b_attached; /* attached inodes, protected by list_lock */
156 struct list_head offline_node; /* anchored at offline_cgwbs */
159 struct work_struct release_work;
165 struct backing_dev_info {
167 struct rb_node rb_node; /* keyed by ->id */
168 struct list_head bdi_list;
169 unsigned long ra_pages; /* max readahead in PAGE_SIZE units */
170 unsigned long io_pages; /* max allowed IO size */
172 struct kref refcnt; /* Reference counter for the structure */
173 unsigned int capabilities; /* Device capabilities */
174 unsigned int min_ratio;
175 unsigned int max_ratio, max_prop_frac;
178 * Sum of avg_write_bw of wbs with dirty inodes. > 0 if there are
179 * any dirty wbs, which is depended upon by bdi_has_dirty().
181 atomic_long_t tot_write_bandwidth;
183 struct bdi_writeback wb; /* the root writeback info for this bdi */
184 struct list_head wb_list; /* list of all wbs */
185 #ifdef CONFIG_CGROUP_WRITEBACK
186 struct radix_tree_root cgwb_tree; /* radix tree of active cgroup wbs */
187 struct mutex cgwb_release_mutex; /* protect shutdown of wb structs */
188 struct rw_semaphore wb_switch_rwsem; /* no cgwb switch while syncing */
190 wait_queue_head_t wb_waitq;
194 struct device *owner;
196 struct timer_list laptop_mode_wb_timer;
198 #ifdef CONFIG_DEBUG_FS
199 struct dentry *debug_dir;
203 struct wb_lock_cookie {
208 #ifdef CONFIG_CGROUP_WRITEBACK
211 * wb_tryget - try to increment a wb's refcount
212 * @wb: bdi_writeback to get
214 static inline bool wb_tryget(struct bdi_writeback *wb)
216 if (wb != &wb->bdi->wb)
217 return percpu_ref_tryget(&wb->refcnt);
222 * wb_get - increment a wb's refcount
223 * @wb: bdi_writeback to get
225 static inline void wb_get(struct bdi_writeback *wb)
227 if (wb != &wb->bdi->wb)
228 percpu_ref_get(&wb->refcnt);
232 * wb_put - decrement a wb's refcount
233 * @wb: bdi_writeback to put
234 * @nr: number of references to put
236 static inline void wb_put_many(struct bdi_writeback *wb, unsigned long nr)
238 if (WARN_ON_ONCE(!wb->bdi)) {
240 * A driver bug might cause a file to be removed before bdi was
246 if (wb != &wb->bdi->wb)
247 percpu_ref_put_many(&wb->refcnt, nr);
251 * wb_put - decrement a wb's refcount
252 * @wb: bdi_writeback to put
254 static inline void wb_put(struct bdi_writeback *wb)
260 * wb_dying - is a wb dying?
261 * @wb: bdi_writeback of interest
263 * Returns whether @wb is unlinked and being drained.
265 static inline bool wb_dying(struct bdi_writeback *wb)
267 return percpu_ref_is_dying(&wb->refcnt);
270 #else /* CONFIG_CGROUP_WRITEBACK */
272 static inline bool wb_tryget(struct bdi_writeback *wb)
277 static inline void wb_get(struct bdi_writeback *wb)
281 static inline void wb_put(struct bdi_writeback *wb)
285 static inline void wb_put_many(struct bdi_writeback *wb, unsigned long nr)
289 static inline bool wb_dying(struct bdi_writeback *wb)
294 #endif /* CONFIG_CGROUP_WRITEBACK */
296 #endif /* __LINUX_BACKING_DEV_DEFS_H */