xref: /linux-6.15/include/linux/backing-dev.h (revision ae82291e)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * include/linux/backing-dev.h
4  *
5  * low-level device information and state which is propagated up through
6  * to high-level code.
7  */
8 
9 #ifndef _LINUX_BACKING_DEV_H
10 #define _LINUX_BACKING_DEV_H
11 
12 #include <linux/kernel.h>
13 #include <linux/fs.h>
14 #include <linux/sched.h>
15 #include <linux/device.h>
16 #include <linux/writeback.h>
17 #include <linux/backing-dev-defs.h>
18 #include <linux/slab.h>
19 
20 static inline struct backing_dev_info *bdi_get(struct backing_dev_info *bdi)
21 {
22 	kref_get(&bdi->refcnt);
23 	return bdi;
24 }
25 
26 struct backing_dev_info *bdi_get_by_id(u64 id);
27 void bdi_put(struct backing_dev_info *bdi);
28 
29 __printf(2, 3)
30 int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...);
31 __printf(2, 0)
32 int bdi_register_va(struct backing_dev_info *bdi, const char *fmt,
33 		    va_list args);
34 void bdi_set_owner(struct backing_dev_info *bdi, struct device *owner);
35 void bdi_unregister(struct backing_dev_info *bdi);
36 
37 struct backing_dev_info *bdi_alloc(int node_id);
38 
39 void wb_start_background_writeback(struct bdi_writeback *wb);
40 void wb_workfn(struct work_struct *work);
41 void wb_wakeup_delayed(struct bdi_writeback *wb);
42 
43 void wb_wait_for_completion(struct wb_completion *done);
44 
45 extern spinlock_t bdi_lock;
46 extern struct list_head bdi_list;
47 
48 extern struct workqueue_struct *bdi_wq;
49 extern struct workqueue_struct *bdi_async_bio_wq;
50 
51 static inline bool wb_has_dirty_io(struct bdi_writeback *wb)
52 {
53 	return test_bit(WB_has_dirty_io, &wb->state);
54 }
55 
56 static inline bool bdi_has_dirty_io(struct backing_dev_info *bdi)
57 {
58 	/*
59 	 * @bdi->tot_write_bandwidth is guaranteed to be > 0 if there are
60 	 * any dirty wbs.  See wb_update_write_bandwidth().
61 	 */
62 	return atomic_long_read(&bdi->tot_write_bandwidth);
63 }
64 
65 static inline void wb_stat_mod(struct bdi_writeback *wb,
66 				 enum wb_stat_item item, s64 amount)
67 {
68 	percpu_counter_add_batch(&wb->stat[item], amount, WB_STAT_BATCH);
69 }
70 
71 static inline void inc_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
72 {
73 	wb_stat_mod(wb, item, 1);
74 }
75 
76 static inline void dec_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
77 {
78 	wb_stat_mod(wb, item, -1);
79 }
80 
81 static inline s64 wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
82 {
83 	return percpu_counter_read_positive(&wb->stat[item]);
84 }
85 
86 static inline s64 wb_stat_sum(struct bdi_writeback *wb, enum wb_stat_item item)
87 {
88 	return percpu_counter_sum_positive(&wb->stat[item]);
89 }
90 
91 extern void wb_writeout_inc(struct bdi_writeback *wb);
92 
93 /*
94  * maximal error of a stat counter.
95  */
96 static inline unsigned long wb_stat_error(void)
97 {
98 #ifdef CONFIG_SMP
99 	return nr_cpu_ids * WB_STAT_BATCH;
100 #else
101 	return 1;
102 #endif
103 }
104 
105 /* BDI ratio is expressed as part per 1000000 for finer granularity. */
106 #define BDI_RATIO_SCALE 10000
107 
108 int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
109 int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
110 int bdi_set_strict_limit(struct backing_dev_info *bdi, unsigned int strict_limit);
111 
112 /*
113  * Flags in backing_dev_info::capability
114  *
115  * BDI_CAP_WRITEBACK:		Supports dirty page writeback, and dirty pages
116  *				should contribute to accounting
117  * BDI_CAP_WRITEBACK_ACCT:	Automatically account writeback pages
118  * BDI_CAP_STRICTLIMIT:		Keep number of dirty pages below bdi threshold
119  */
120 #define BDI_CAP_WRITEBACK		(1 << 0)
121 #define BDI_CAP_WRITEBACK_ACCT		(1 << 1)
122 #define BDI_CAP_STRICTLIMIT		(1 << 2)
123 
124 extern struct backing_dev_info noop_backing_dev_info;
125 
126 int bdi_init(struct backing_dev_info *bdi);
127 
128 /**
129  * writeback_in_progress - determine whether there is writeback in progress
130  * @wb: bdi_writeback of interest
131  *
132  * Determine whether there is writeback waiting to be handled against a
133  * bdi_writeback.
134  */
135 static inline bool writeback_in_progress(struct bdi_writeback *wb)
136 {
137 	return test_bit(WB_writeback_running, &wb->state);
138 }
139 
140 struct backing_dev_info *inode_to_bdi(struct inode *inode);
141 
142 static inline bool mapping_can_writeback(struct address_space *mapping)
143 {
144 	return inode_to_bdi(mapping->host)->capabilities & BDI_CAP_WRITEBACK;
145 }
146 
147 #ifdef CONFIG_CGROUP_WRITEBACK
148 
149 struct bdi_writeback *wb_get_lookup(struct backing_dev_info *bdi,
150 				    struct cgroup_subsys_state *memcg_css);
151 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
152 				    struct cgroup_subsys_state *memcg_css,
153 				    gfp_t gfp);
154 void wb_memcg_offline(struct mem_cgroup *memcg);
155 void wb_blkcg_offline(struct cgroup_subsys_state *css);
156 
157 /**
158  * inode_cgwb_enabled - test whether cgroup writeback is enabled on an inode
159  * @inode: inode of interest
160  *
161  * Cgroup writeback requires support from the filesystem.  Also, both memcg and
162  * iocg have to be on the default hierarchy.  Test whether all conditions are
163  * met.
164  *
165  * Note that the test result may change dynamically on the same inode
166  * depending on how memcg and iocg are configured.
167  */
168 static inline bool inode_cgwb_enabled(struct inode *inode)
169 {
170 	struct backing_dev_info *bdi = inode_to_bdi(inode);
171 
172 	return cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
173 		cgroup_subsys_on_dfl(io_cgrp_subsys) &&
174 		(bdi->capabilities & BDI_CAP_WRITEBACK) &&
175 		(inode->i_sb->s_iflags & SB_I_CGROUPWB);
176 }
177 
178 /**
179  * wb_find_current - find wb for %current on a bdi
180  * @bdi: bdi of interest
181  *
182  * Find the wb of @bdi which matches both the memcg and blkcg of %current.
183  * Must be called under rcu_read_lock() which protects the returend wb.
184  * NULL if not found.
185  */
186 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
187 {
188 	struct cgroup_subsys_state *memcg_css;
189 	struct bdi_writeback *wb;
190 
191 	memcg_css = task_css(current, memory_cgrp_id);
192 	if (!memcg_css->parent)
193 		return &bdi->wb;
194 
195 	wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
196 
197 	/*
198 	 * %current's blkcg equals the effective blkcg of its memcg.  No
199 	 * need to use the relatively expensive cgroup_get_e_css().
200 	 */
201 	if (likely(wb && wb->blkcg_css == task_css(current, io_cgrp_id)))
202 		return wb;
203 	return NULL;
204 }
205 
206 /**
207  * wb_get_create_current - get or create wb for %current on a bdi
208  * @bdi: bdi of interest
209  * @gfp: allocation mask
210  *
211  * Equivalent to wb_get_create() on %current's memcg.  This function is
212  * called from a relatively hot path and optimizes the common cases using
213  * wb_find_current().
214  */
215 static inline struct bdi_writeback *
216 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
217 {
218 	struct bdi_writeback *wb;
219 
220 	rcu_read_lock();
221 	wb = wb_find_current(bdi);
222 	if (wb && unlikely(!wb_tryget(wb)))
223 		wb = NULL;
224 	rcu_read_unlock();
225 
226 	if (unlikely(!wb)) {
227 		struct cgroup_subsys_state *memcg_css;
228 
229 		memcg_css = task_get_css(current, memory_cgrp_id);
230 		wb = wb_get_create(bdi, memcg_css, gfp);
231 		css_put(memcg_css);
232 	}
233 	return wb;
234 }
235 
236 /**
237  * inode_to_wb - determine the wb of an inode
238  * @inode: inode of interest
239  *
240  * Returns the wb @inode is currently associated with.  The caller must be
241  * holding either @inode->i_lock, the i_pages lock, or the
242  * associated wb's list_lock.
243  */
244 static inline struct bdi_writeback *inode_to_wb(const struct inode *inode)
245 {
246 #ifdef CONFIG_LOCKDEP
247 	WARN_ON_ONCE(debug_locks &&
248 		     (!lockdep_is_held(&inode->i_lock) &&
249 		      !lockdep_is_held(&inode->i_mapping->i_pages.xa_lock) &&
250 		      !lockdep_is_held(&inode->i_wb->list_lock)));
251 #endif
252 	return inode->i_wb;
253 }
254 
255 static inline struct bdi_writeback *inode_to_wb_wbc(
256 				struct inode *inode,
257 				struct writeback_control *wbc)
258 {
259 	/*
260 	 * If wbc does not have inode attached, it means cgroup writeback was
261 	 * disabled when wbc started. Just use the default wb in that case.
262 	 */
263 	return wbc->wb ? wbc->wb : &inode_to_bdi(inode)->wb;
264 }
265 
266 /**
267  * unlocked_inode_to_wb_begin - begin unlocked inode wb access transaction
268  * @inode: target inode
269  * @cookie: output param, to be passed to the end function
270  *
271  * The caller wants to access the wb associated with @inode but isn't
272  * holding inode->i_lock, the i_pages lock or wb->list_lock.  This
273  * function determines the wb associated with @inode and ensures that the
274  * association doesn't change until the transaction is finished with
275  * unlocked_inode_to_wb_end().
276  *
277  * The caller must call unlocked_inode_to_wb_end() with *@cookie afterwards and
278  * can't sleep during the transaction.  IRQs may or may not be disabled on
279  * return.
280  */
281 static inline struct bdi_writeback *
282 unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie)
283 {
284 	rcu_read_lock();
285 
286 	/*
287 	 * Paired with store_release in inode_switch_wbs_work_fn() and
288 	 * ensures that we see the new wb if we see cleared I_WB_SWITCH.
289 	 */
290 	cookie->locked = smp_load_acquire(&inode->i_state) & I_WB_SWITCH;
291 
292 	if (unlikely(cookie->locked))
293 		xa_lock_irqsave(&inode->i_mapping->i_pages, cookie->flags);
294 
295 	/*
296 	 * Protected by either !I_WB_SWITCH + rcu_read_lock() or the i_pages
297 	 * lock.  inode_to_wb() will bark.  Deref directly.
298 	 */
299 	return inode->i_wb;
300 }
301 
302 /**
303  * unlocked_inode_to_wb_end - end inode wb access transaction
304  * @inode: target inode
305  * @cookie: @cookie from unlocked_inode_to_wb_begin()
306  */
307 static inline void unlocked_inode_to_wb_end(struct inode *inode,
308 					    struct wb_lock_cookie *cookie)
309 {
310 	if (unlikely(cookie->locked))
311 		xa_unlock_irqrestore(&inode->i_mapping->i_pages, cookie->flags);
312 
313 	rcu_read_unlock();
314 }
315 
316 #else	/* CONFIG_CGROUP_WRITEBACK */
317 
318 static inline bool inode_cgwb_enabled(struct inode *inode)
319 {
320 	return false;
321 }
322 
323 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
324 {
325 	return &bdi->wb;
326 }
327 
328 static inline struct bdi_writeback *
329 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
330 {
331 	return &bdi->wb;
332 }
333 
334 static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
335 {
336 	return &inode_to_bdi(inode)->wb;
337 }
338 
339 static inline struct bdi_writeback *inode_to_wb_wbc(
340 				struct inode *inode,
341 				struct writeback_control *wbc)
342 {
343 	return inode_to_wb(inode);
344 }
345 
346 
347 static inline struct bdi_writeback *
348 unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie)
349 {
350 	return inode_to_wb(inode);
351 }
352 
353 static inline void unlocked_inode_to_wb_end(struct inode *inode,
354 					    struct wb_lock_cookie *cookie)
355 {
356 }
357 
358 static inline void wb_memcg_offline(struct mem_cgroup *memcg)
359 {
360 }
361 
362 static inline void wb_blkcg_offline(struct cgroup_subsys_state *css)
363 {
364 }
365 
366 #endif	/* CONFIG_CGROUP_WRITEBACK */
367 
368 const char *bdi_dev_name(struct backing_dev_info *bdi);
369 
370 #endif	/* _LINUX_BACKING_DEV_H */
371