xref: /linux-6.15/include/linux/backing-dev.h (revision b9b1335e)
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 struct blkcg;
21 
22 static inline struct backing_dev_info *bdi_get(struct backing_dev_info *bdi)
23 {
24 	kref_get(&bdi->refcnt);
25 	return bdi;
26 }
27 
28 struct backing_dev_info *bdi_get_by_id(u64 id);
29 void bdi_put(struct backing_dev_info *bdi);
30 
31 __printf(2, 3)
32 int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...);
33 __printf(2, 0)
34 int bdi_register_va(struct backing_dev_info *bdi, const char *fmt,
35 		    va_list args);
36 void bdi_set_owner(struct backing_dev_info *bdi, struct device *owner);
37 void bdi_unregister(struct backing_dev_info *bdi);
38 
39 struct backing_dev_info *bdi_alloc(int node_id);
40 
41 void wb_start_background_writeback(struct bdi_writeback *wb);
42 void wb_workfn(struct work_struct *work);
43 void wb_wakeup_delayed(struct bdi_writeback *wb);
44 
45 void wb_wait_for_completion(struct wb_completion *done);
46 
47 extern spinlock_t bdi_lock;
48 extern struct list_head bdi_list;
49 
50 extern struct workqueue_struct *bdi_wq;
51 extern struct workqueue_struct *bdi_async_bio_wq;
52 
53 static inline bool wb_has_dirty_io(struct bdi_writeback *wb)
54 {
55 	return test_bit(WB_has_dirty_io, &wb->state);
56 }
57 
58 static inline bool bdi_has_dirty_io(struct backing_dev_info *bdi)
59 {
60 	/*
61 	 * @bdi->tot_write_bandwidth is guaranteed to be > 0 if there are
62 	 * any dirty wbs.  See wb_update_write_bandwidth().
63 	 */
64 	return atomic_long_read(&bdi->tot_write_bandwidth);
65 }
66 
67 static inline void wb_stat_mod(struct bdi_writeback *wb,
68 				 enum wb_stat_item item, s64 amount)
69 {
70 	percpu_counter_add_batch(&wb->stat[item], amount, WB_STAT_BATCH);
71 }
72 
73 static inline void inc_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
74 {
75 	wb_stat_mod(wb, item, 1);
76 }
77 
78 static inline void dec_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
79 {
80 	wb_stat_mod(wb, item, -1);
81 }
82 
83 static inline s64 wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
84 {
85 	return percpu_counter_read_positive(&wb->stat[item]);
86 }
87 
88 static inline s64 wb_stat_sum(struct bdi_writeback *wb, enum wb_stat_item item)
89 {
90 	return percpu_counter_sum_positive(&wb->stat[item]);
91 }
92 
93 extern void wb_writeout_inc(struct bdi_writeback *wb);
94 
95 /*
96  * maximal error of a stat counter.
97  */
98 static inline unsigned long wb_stat_error(void)
99 {
100 #ifdef CONFIG_SMP
101 	return nr_cpu_ids * WB_STAT_BATCH;
102 #else
103 	return 1;
104 #endif
105 }
106 
107 int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
108 int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
109 
110 /*
111  * Flags in backing_dev_info::capability
112  *
113  * BDI_CAP_WRITEBACK:		Supports dirty page writeback, and dirty pages
114  *				should contribute to accounting
115  * BDI_CAP_WRITEBACK_ACCT:	Automatically account writeback pages
116  * BDI_CAP_STRICTLIMIT:		Keep number of dirty pages below bdi threshold
117  */
118 #define BDI_CAP_WRITEBACK		(1 << 0)
119 #define BDI_CAP_WRITEBACK_ACCT		(1 << 1)
120 #define BDI_CAP_STRICTLIMIT		(1 << 2)
121 
122 extern struct backing_dev_info noop_backing_dev_info;
123 
124 /**
125  * writeback_in_progress - determine whether there is writeback in progress
126  * @wb: bdi_writeback of interest
127  *
128  * Determine whether there is writeback waiting to be handled against a
129  * bdi_writeback.
130  */
131 static inline bool writeback_in_progress(struct bdi_writeback *wb)
132 {
133 	return test_bit(WB_writeback_running, &wb->state);
134 }
135 
136 struct backing_dev_info *inode_to_bdi(struct inode *inode);
137 
138 long congestion_wait(int sync, long timeout);
139 
140 static inline bool mapping_can_writeback(struct address_space *mapping)
141 {
142 	return inode_to_bdi(mapping->host)->capabilities & BDI_CAP_WRITEBACK;
143 }
144 
145 static inline int bdi_sched_wait(void *word)
146 {
147 	schedule();
148 	return 0;
149 }
150 
151 #ifdef CONFIG_CGROUP_WRITEBACK
152 
153 struct bdi_writeback *wb_get_lookup(struct backing_dev_info *bdi,
154 				    struct cgroup_subsys_state *memcg_css);
155 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
156 				    struct cgroup_subsys_state *memcg_css,
157 				    gfp_t gfp);
158 void wb_memcg_offline(struct mem_cgroup *memcg);
159 void wb_blkcg_offline(struct blkcg *blkcg);
160 
161 /**
162  * inode_cgwb_enabled - test whether cgroup writeback is enabled on an inode
163  * @inode: inode of interest
164  *
165  * Cgroup writeback requires support from the filesystem.  Also, both memcg and
166  * iocg have to be on the default hierarchy.  Test whether all conditions are
167  * met.
168  *
169  * Note that the test result may change dynamically on the same inode
170  * depending on how memcg and iocg are configured.
171  */
172 static inline bool inode_cgwb_enabled(struct inode *inode)
173 {
174 	struct backing_dev_info *bdi = inode_to_bdi(inode);
175 
176 	return cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
177 		cgroup_subsys_on_dfl(io_cgrp_subsys) &&
178 		(bdi->capabilities & BDI_CAP_WRITEBACK) &&
179 		(inode->i_sb->s_iflags & SB_I_CGROUPWB);
180 }
181 
182 /**
183  * wb_find_current - find wb for %current on a bdi
184  * @bdi: bdi of interest
185  *
186  * Find the wb of @bdi which matches both the memcg and blkcg of %current.
187  * Must be called under rcu_read_lock() which protects the returend wb.
188  * NULL if not found.
189  */
190 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
191 {
192 	struct cgroup_subsys_state *memcg_css;
193 	struct bdi_writeback *wb;
194 
195 	memcg_css = task_css(current, memory_cgrp_id);
196 	if (!memcg_css->parent)
197 		return &bdi->wb;
198 
199 	wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
200 
201 	/*
202 	 * %current's blkcg equals the effective blkcg of its memcg.  No
203 	 * need to use the relatively expensive cgroup_get_e_css().
204 	 */
205 	if (likely(wb && wb->blkcg_css == task_css(current, io_cgrp_id)))
206 		return wb;
207 	return NULL;
208 }
209 
210 /**
211  * wb_get_create_current - get or create wb for %current on a bdi
212  * @bdi: bdi of interest
213  * @gfp: allocation mask
214  *
215  * Equivalent to wb_get_create() on %current's memcg.  This function is
216  * called from a relatively hot path and optimizes the common cases using
217  * wb_find_current().
218  */
219 static inline struct bdi_writeback *
220 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
221 {
222 	struct bdi_writeback *wb;
223 
224 	rcu_read_lock();
225 	wb = wb_find_current(bdi);
226 	if (wb && unlikely(!wb_tryget(wb)))
227 		wb = NULL;
228 	rcu_read_unlock();
229 
230 	if (unlikely(!wb)) {
231 		struct cgroup_subsys_state *memcg_css;
232 
233 		memcg_css = task_get_css(current, memory_cgrp_id);
234 		wb = wb_get_create(bdi, memcg_css, gfp);
235 		css_put(memcg_css);
236 	}
237 	return wb;
238 }
239 
240 /**
241  * inode_to_wb_is_valid - test whether an inode has a wb associated
242  * @inode: inode of interest
243  *
244  * Returns %true if @inode has a wb associated.  May be called without any
245  * locking.
246  */
247 static inline bool inode_to_wb_is_valid(struct inode *inode)
248 {
249 	return inode->i_wb;
250 }
251 
252 /**
253  * inode_to_wb - determine the wb of an inode
254  * @inode: inode of interest
255  *
256  * Returns the wb @inode is currently associated with.  The caller must be
257  * holding either @inode->i_lock, the i_pages lock, or the
258  * associated wb's list_lock.
259  */
260 static inline struct bdi_writeback *inode_to_wb(const struct inode *inode)
261 {
262 #ifdef CONFIG_LOCKDEP
263 	WARN_ON_ONCE(debug_locks &&
264 		     (!lockdep_is_held(&inode->i_lock) &&
265 		      !lockdep_is_held(&inode->i_mapping->i_pages.xa_lock) &&
266 		      !lockdep_is_held(&inode->i_wb->list_lock)));
267 #endif
268 	return inode->i_wb;
269 }
270 
271 static inline struct bdi_writeback *inode_to_wb_wbc(
272 				struct inode *inode,
273 				struct writeback_control *wbc)
274 {
275 	/*
276 	 * If wbc does not have inode attached, it means cgroup writeback was
277 	 * disabled when wbc started. Just use the default wb in that case.
278 	 */
279 	return wbc->wb ? wbc->wb : &inode_to_bdi(inode)->wb;
280 }
281 
282 /**
283  * unlocked_inode_to_wb_begin - begin unlocked inode wb access transaction
284  * @inode: target inode
285  * @cookie: output param, to be passed to the end function
286  *
287  * The caller wants to access the wb associated with @inode but isn't
288  * holding inode->i_lock, the i_pages lock or wb->list_lock.  This
289  * function determines the wb associated with @inode and ensures that the
290  * association doesn't change until the transaction is finished with
291  * unlocked_inode_to_wb_end().
292  *
293  * The caller must call unlocked_inode_to_wb_end() with *@cookie afterwards and
294  * can't sleep during the transaction.  IRQs may or may not be disabled on
295  * return.
296  */
297 static inline struct bdi_writeback *
298 unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie)
299 {
300 	rcu_read_lock();
301 
302 	/*
303 	 * Paired with store_release in inode_switch_wbs_work_fn() and
304 	 * ensures that we see the new wb if we see cleared I_WB_SWITCH.
305 	 */
306 	cookie->locked = smp_load_acquire(&inode->i_state) & I_WB_SWITCH;
307 
308 	if (unlikely(cookie->locked))
309 		xa_lock_irqsave(&inode->i_mapping->i_pages, cookie->flags);
310 
311 	/*
312 	 * Protected by either !I_WB_SWITCH + rcu_read_lock() or the i_pages
313 	 * lock.  inode_to_wb() will bark.  Deref directly.
314 	 */
315 	return inode->i_wb;
316 }
317 
318 /**
319  * unlocked_inode_to_wb_end - end inode wb access transaction
320  * @inode: target inode
321  * @cookie: @cookie from unlocked_inode_to_wb_begin()
322  */
323 static inline void unlocked_inode_to_wb_end(struct inode *inode,
324 					    struct wb_lock_cookie *cookie)
325 {
326 	if (unlikely(cookie->locked))
327 		xa_unlock_irqrestore(&inode->i_mapping->i_pages, cookie->flags);
328 
329 	rcu_read_unlock();
330 }
331 
332 #else	/* CONFIG_CGROUP_WRITEBACK */
333 
334 static inline bool inode_cgwb_enabled(struct inode *inode)
335 {
336 	return false;
337 }
338 
339 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
340 {
341 	return &bdi->wb;
342 }
343 
344 static inline struct bdi_writeback *
345 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
346 {
347 	return &bdi->wb;
348 }
349 
350 static inline bool inode_to_wb_is_valid(struct inode *inode)
351 {
352 	return true;
353 }
354 
355 static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
356 {
357 	return &inode_to_bdi(inode)->wb;
358 }
359 
360 static inline struct bdi_writeback *inode_to_wb_wbc(
361 				struct inode *inode,
362 				struct writeback_control *wbc)
363 {
364 	return inode_to_wb(inode);
365 }
366 
367 
368 static inline struct bdi_writeback *
369 unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie)
370 {
371 	return inode_to_wb(inode);
372 }
373 
374 static inline void unlocked_inode_to_wb_end(struct inode *inode,
375 					    struct wb_lock_cookie *cookie)
376 {
377 }
378 
379 static inline void wb_memcg_offline(struct mem_cgroup *memcg)
380 {
381 }
382 
383 static inline void wb_blkcg_offline(struct blkcg *blkcg)
384 {
385 }
386 
387 #endif	/* CONFIG_CGROUP_WRITEBACK */
388 
389 const char *bdi_dev_name(struct backing_dev_info *bdi);
390 
391 #endif	/* _LINUX_BACKING_DEV_H */
392