xref: /linux-6.15/include/linux/backing-dev.h (revision 712c00d6)
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 u64 bdi_get_min_bytes(struct backing_dev_info *bdi);
109 u64 bdi_get_max_bytes(struct backing_dev_info *bdi);
110 int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
111 int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
112 int bdi_set_max_bytes(struct backing_dev_info *bdi, u64 max_bytes);
113 int bdi_set_strict_limit(struct backing_dev_info *bdi, unsigned int strict_limit);
114 
115 /*
116  * Flags in backing_dev_info::capability
117  *
118  * BDI_CAP_WRITEBACK:		Supports dirty page writeback, and dirty pages
119  *				should contribute to accounting
120  * BDI_CAP_WRITEBACK_ACCT:	Automatically account writeback pages
121  * BDI_CAP_STRICTLIMIT:		Keep number of dirty pages below bdi threshold
122  */
123 #define BDI_CAP_WRITEBACK		(1 << 0)
124 #define BDI_CAP_WRITEBACK_ACCT		(1 << 1)
125 #define BDI_CAP_STRICTLIMIT		(1 << 2)
126 
127 extern struct backing_dev_info noop_backing_dev_info;
128 
129 int bdi_init(struct backing_dev_info *bdi);
130 
131 /**
132  * writeback_in_progress - determine whether there is writeback in progress
133  * @wb: bdi_writeback of interest
134  *
135  * Determine whether there is writeback waiting to be handled against a
136  * bdi_writeback.
137  */
138 static inline bool writeback_in_progress(struct bdi_writeback *wb)
139 {
140 	return test_bit(WB_writeback_running, &wb->state);
141 }
142 
143 struct backing_dev_info *inode_to_bdi(struct inode *inode);
144 
145 static inline bool mapping_can_writeback(struct address_space *mapping)
146 {
147 	return inode_to_bdi(mapping->host)->capabilities & BDI_CAP_WRITEBACK;
148 }
149 
150 #ifdef CONFIG_CGROUP_WRITEBACK
151 
152 struct bdi_writeback *wb_get_lookup(struct backing_dev_info *bdi,
153 				    struct cgroup_subsys_state *memcg_css);
154 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
155 				    struct cgroup_subsys_state *memcg_css,
156 				    gfp_t gfp);
157 void wb_memcg_offline(struct mem_cgroup *memcg);
158 void wb_blkcg_offline(struct cgroup_subsys_state *css);
159 
160 /**
161  * inode_cgwb_enabled - test whether cgroup writeback is enabled on an inode
162  * @inode: inode of interest
163  *
164  * Cgroup writeback requires support from the filesystem.  Also, both memcg and
165  * iocg have to be on the default hierarchy.  Test whether all conditions are
166  * met.
167  *
168  * Note that the test result may change dynamically on the same inode
169  * depending on how memcg and iocg are configured.
170  */
171 static inline bool inode_cgwb_enabled(struct inode *inode)
172 {
173 	struct backing_dev_info *bdi = inode_to_bdi(inode);
174 
175 	return cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
176 		cgroup_subsys_on_dfl(io_cgrp_subsys) &&
177 		(bdi->capabilities & BDI_CAP_WRITEBACK) &&
178 		(inode->i_sb->s_iflags & SB_I_CGROUPWB);
179 }
180 
181 /**
182  * wb_find_current - find wb for %current on a bdi
183  * @bdi: bdi of interest
184  *
185  * Find the wb of @bdi which matches both the memcg and blkcg of %current.
186  * Must be called under rcu_read_lock() which protects the returend wb.
187  * NULL if not found.
188  */
189 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
190 {
191 	struct cgroup_subsys_state *memcg_css;
192 	struct bdi_writeback *wb;
193 
194 	memcg_css = task_css(current, memory_cgrp_id);
195 	if (!memcg_css->parent)
196 		return &bdi->wb;
197 
198 	wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
199 
200 	/*
201 	 * %current's blkcg equals the effective blkcg of its memcg.  No
202 	 * need to use the relatively expensive cgroup_get_e_css().
203 	 */
204 	if (likely(wb && wb->blkcg_css == task_css(current, io_cgrp_id)))
205 		return wb;
206 	return NULL;
207 }
208 
209 /**
210  * wb_get_create_current - get or create wb for %current on a bdi
211  * @bdi: bdi of interest
212  * @gfp: allocation mask
213  *
214  * Equivalent to wb_get_create() on %current's memcg.  This function is
215  * called from a relatively hot path and optimizes the common cases using
216  * wb_find_current().
217  */
218 static inline struct bdi_writeback *
219 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
220 {
221 	struct bdi_writeback *wb;
222 
223 	rcu_read_lock();
224 	wb = wb_find_current(bdi);
225 	if (wb && unlikely(!wb_tryget(wb)))
226 		wb = NULL;
227 	rcu_read_unlock();
228 
229 	if (unlikely(!wb)) {
230 		struct cgroup_subsys_state *memcg_css;
231 
232 		memcg_css = task_get_css(current, memory_cgrp_id);
233 		wb = wb_get_create(bdi, memcg_css, gfp);
234 		css_put(memcg_css);
235 	}
236 	return wb;
237 }
238 
239 /**
240  * inode_to_wb - determine the wb of an inode
241  * @inode: inode of interest
242  *
243  * Returns the wb @inode is currently associated with.  The caller must be
244  * holding either @inode->i_lock, the i_pages lock, or the
245  * associated wb's list_lock.
246  */
247 static inline struct bdi_writeback *inode_to_wb(const struct inode *inode)
248 {
249 #ifdef CONFIG_LOCKDEP
250 	WARN_ON_ONCE(debug_locks &&
251 		     (!lockdep_is_held(&inode->i_lock) &&
252 		      !lockdep_is_held(&inode->i_mapping->i_pages.xa_lock) &&
253 		      !lockdep_is_held(&inode->i_wb->list_lock)));
254 #endif
255 	return inode->i_wb;
256 }
257 
258 static inline struct bdi_writeback *inode_to_wb_wbc(
259 				struct inode *inode,
260 				struct writeback_control *wbc)
261 {
262 	/*
263 	 * If wbc does not have inode attached, it means cgroup writeback was
264 	 * disabled when wbc started. Just use the default wb in that case.
265 	 */
266 	return wbc->wb ? wbc->wb : &inode_to_bdi(inode)->wb;
267 }
268 
269 /**
270  * unlocked_inode_to_wb_begin - begin unlocked inode wb access transaction
271  * @inode: target inode
272  * @cookie: output param, to be passed to the end function
273  *
274  * The caller wants to access the wb associated with @inode but isn't
275  * holding inode->i_lock, the i_pages lock or wb->list_lock.  This
276  * function determines the wb associated with @inode and ensures that the
277  * association doesn't change until the transaction is finished with
278  * unlocked_inode_to_wb_end().
279  *
280  * The caller must call unlocked_inode_to_wb_end() with *@cookie afterwards and
281  * can't sleep during the transaction.  IRQs may or may not be disabled on
282  * return.
283  */
284 static inline struct bdi_writeback *
285 unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie)
286 {
287 	rcu_read_lock();
288 
289 	/*
290 	 * Paired with store_release in inode_switch_wbs_work_fn() and
291 	 * ensures that we see the new wb if we see cleared I_WB_SWITCH.
292 	 */
293 	cookie->locked = smp_load_acquire(&inode->i_state) & I_WB_SWITCH;
294 
295 	if (unlikely(cookie->locked))
296 		xa_lock_irqsave(&inode->i_mapping->i_pages, cookie->flags);
297 
298 	/*
299 	 * Protected by either !I_WB_SWITCH + rcu_read_lock() or the i_pages
300 	 * lock.  inode_to_wb() will bark.  Deref directly.
301 	 */
302 	return inode->i_wb;
303 }
304 
305 /**
306  * unlocked_inode_to_wb_end - end inode wb access transaction
307  * @inode: target inode
308  * @cookie: @cookie from unlocked_inode_to_wb_begin()
309  */
310 static inline void unlocked_inode_to_wb_end(struct inode *inode,
311 					    struct wb_lock_cookie *cookie)
312 {
313 	if (unlikely(cookie->locked))
314 		xa_unlock_irqrestore(&inode->i_mapping->i_pages, cookie->flags);
315 
316 	rcu_read_unlock();
317 }
318 
319 #else	/* CONFIG_CGROUP_WRITEBACK */
320 
321 static inline bool inode_cgwb_enabled(struct inode *inode)
322 {
323 	return false;
324 }
325 
326 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
327 {
328 	return &bdi->wb;
329 }
330 
331 static inline struct bdi_writeback *
332 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
333 {
334 	return &bdi->wb;
335 }
336 
337 static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
338 {
339 	return &inode_to_bdi(inode)->wb;
340 }
341 
342 static inline struct bdi_writeback *inode_to_wb_wbc(
343 				struct inode *inode,
344 				struct writeback_control *wbc)
345 {
346 	return inode_to_wb(inode);
347 }
348 
349 
350 static inline struct bdi_writeback *
351 unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie)
352 {
353 	return inode_to_wb(inode);
354 }
355 
356 static inline void unlocked_inode_to_wb_end(struct inode *inode,
357 					    struct wb_lock_cookie *cookie)
358 {
359 }
360 
361 static inline void wb_memcg_offline(struct mem_cgroup *memcg)
362 {
363 }
364 
365 static inline void wb_blkcg_offline(struct cgroup_subsys_state *css)
366 {
367 }
368 
369 #endif	/* CONFIG_CGROUP_WRITEBACK */
370 
371 const char *bdi_dev_name(struct backing_dev_info *bdi);
372 
373 #endif	/* _LINUX_BACKING_DEV_H */
374