xref: /linux-6.15/include/linux/backing-dev.h (revision 1bf27e98)
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_max_bytes(struct backing_dev_info *bdi);
109 int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
110 int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
111 int bdi_set_max_bytes(struct backing_dev_info *bdi, u64 max_bytes);
112 int bdi_set_strict_limit(struct backing_dev_info *bdi, unsigned int strict_limit);
113 
114 /*
115  * Flags in backing_dev_info::capability
116  *
117  * BDI_CAP_WRITEBACK:		Supports dirty page writeback, and dirty pages
118  *				should contribute to accounting
119  * BDI_CAP_WRITEBACK_ACCT:	Automatically account writeback pages
120  * BDI_CAP_STRICTLIMIT:		Keep number of dirty pages below bdi threshold
121  */
122 #define BDI_CAP_WRITEBACK		(1 << 0)
123 #define BDI_CAP_WRITEBACK_ACCT		(1 << 1)
124 #define BDI_CAP_STRICTLIMIT		(1 << 2)
125 
126 extern struct backing_dev_info noop_backing_dev_info;
127 
128 int bdi_init(struct backing_dev_info *bdi);
129 
130 /**
131  * writeback_in_progress - determine whether there is writeback in progress
132  * @wb: bdi_writeback of interest
133  *
134  * Determine whether there is writeback waiting to be handled against a
135  * bdi_writeback.
136  */
137 static inline bool writeback_in_progress(struct bdi_writeback *wb)
138 {
139 	return test_bit(WB_writeback_running, &wb->state);
140 }
141 
142 struct backing_dev_info *inode_to_bdi(struct inode *inode);
143 
144 static inline bool mapping_can_writeback(struct address_space *mapping)
145 {
146 	return inode_to_bdi(mapping->host)->capabilities & BDI_CAP_WRITEBACK;
147 }
148 
149 #ifdef CONFIG_CGROUP_WRITEBACK
150 
151 struct bdi_writeback *wb_get_lookup(struct backing_dev_info *bdi,
152 				    struct cgroup_subsys_state *memcg_css);
153 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
154 				    struct cgroup_subsys_state *memcg_css,
155 				    gfp_t gfp);
156 void wb_memcg_offline(struct mem_cgroup *memcg);
157 void wb_blkcg_offline(struct cgroup_subsys_state *css);
158 
159 /**
160  * inode_cgwb_enabled - test whether cgroup writeback is enabled on an inode
161  * @inode: inode of interest
162  *
163  * Cgroup writeback requires support from the filesystem.  Also, both memcg and
164  * iocg have to be on the default hierarchy.  Test whether all conditions are
165  * met.
166  *
167  * Note that the test result may change dynamically on the same inode
168  * depending on how memcg and iocg are configured.
169  */
170 static inline bool inode_cgwb_enabled(struct inode *inode)
171 {
172 	struct backing_dev_info *bdi = inode_to_bdi(inode);
173 
174 	return cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
175 		cgroup_subsys_on_dfl(io_cgrp_subsys) &&
176 		(bdi->capabilities & BDI_CAP_WRITEBACK) &&
177 		(inode->i_sb->s_iflags & SB_I_CGROUPWB);
178 }
179 
180 /**
181  * wb_find_current - find wb for %current on a bdi
182  * @bdi: bdi of interest
183  *
184  * Find the wb of @bdi which matches both the memcg and blkcg of %current.
185  * Must be called under rcu_read_lock() which protects the returend wb.
186  * NULL if not found.
187  */
188 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
189 {
190 	struct cgroup_subsys_state *memcg_css;
191 	struct bdi_writeback *wb;
192 
193 	memcg_css = task_css(current, memory_cgrp_id);
194 	if (!memcg_css->parent)
195 		return &bdi->wb;
196 
197 	wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
198 
199 	/*
200 	 * %current's blkcg equals the effective blkcg of its memcg.  No
201 	 * need to use the relatively expensive cgroup_get_e_css().
202 	 */
203 	if (likely(wb && wb->blkcg_css == task_css(current, io_cgrp_id)))
204 		return wb;
205 	return NULL;
206 }
207 
208 /**
209  * wb_get_create_current - get or create wb for %current on a bdi
210  * @bdi: bdi of interest
211  * @gfp: allocation mask
212  *
213  * Equivalent to wb_get_create() on %current's memcg.  This function is
214  * called from a relatively hot path and optimizes the common cases using
215  * wb_find_current().
216  */
217 static inline struct bdi_writeback *
218 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
219 {
220 	struct bdi_writeback *wb;
221 
222 	rcu_read_lock();
223 	wb = wb_find_current(bdi);
224 	if (wb && unlikely(!wb_tryget(wb)))
225 		wb = NULL;
226 	rcu_read_unlock();
227 
228 	if (unlikely(!wb)) {
229 		struct cgroup_subsys_state *memcg_css;
230 
231 		memcg_css = task_get_css(current, memory_cgrp_id);
232 		wb = wb_get_create(bdi, memcg_css, gfp);
233 		css_put(memcg_css);
234 	}
235 	return wb;
236 }
237 
238 /**
239  * inode_to_wb - determine the wb of an inode
240  * @inode: inode of interest
241  *
242  * Returns the wb @inode is currently associated with.  The caller must be
243  * holding either @inode->i_lock, the i_pages lock, or the
244  * associated wb's list_lock.
245  */
246 static inline struct bdi_writeback *inode_to_wb(const struct inode *inode)
247 {
248 #ifdef CONFIG_LOCKDEP
249 	WARN_ON_ONCE(debug_locks &&
250 		     (!lockdep_is_held(&inode->i_lock) &&
251 		      !lockdep_is_held(&inode->i_mapping->i_pages.xa_lock) &&
252 		      !lockdep_is_held(&inode->i_wb->list_lock)));
253 #endif
254 	return inode->i_wb;
255 }
256 
257 static inline struct bdi_writeback *inode_to_wb_wbc(
258 				struct inode *inode,
259 				struct writeback_control *wbc)
260 {
261 	/*
262 	 * If wbc does not have inode attached, it means cgroup writeback was
263 	 * disabled when wbc started. Just use the default wb in that case.
264 	 */
265 	return wbc->wb ? wbc->wb : &inode_to_bdi(inode)->wb;
266 }
267 
268 /**
269  * unlocked_inode_to_wb_begin - begin unlocked inode wb access transaction
270  * @inode: target inode
271  * @cookie: output param, to be passed to the end function
272  *
273  * The caller wants to access the wb associated with @inode but isn't
274  * holding inode->i_lock, the i_pages lock or wb->list_lock.  This
275  * function determines the wb associated with @inode and ensures that the
276  * association doesn't change until the transaction is finished with
277  * unlocked_inode_to_wb_end().
278  *
279  * The caller must call unlocked_inode_to_wb_end() with *@cookie afterwards and
280  * can't sleep during the transaction.  IRQs may or may not be disabled on
281  * return.
282  */
283 static inline struct bdi_writeback *
284 unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie)
285 {
286 	rcu_read_lock();
287 
288 	/*
289 	 * Paired with store_release in inode_switch_wbs_work_fn() and
290 	 * ensures that we see the new wb if we see cleared I_WB_SWITCH.
291 	 */
292 	cookie->locked = smp_load_acquire(&inode->i_state) & I_WB_SWITCH;
293 
294 	if (unlikely(cookie->locked))
295 		xa_lock_irqsave(&inode->i_mapping->i_pages, cookie->flags);
296 
297 	/*
298 	 * Protected by either !I_WB_SWITCH + rcu_read_lock() or the i_pages
299 	 * lock.  inode_to_wb() will bark.  Deref directly.
300 	 */
301 	return inode->i_wb;
302 }
303 
304 /**
305  * unlocked_inode_to_wb_end - end inode wb access transaction
306  * @inode: target inode
307  * @cookie: @cookie from unlocked_inode_to_wb_begin()
308  */
309 static inline void unlocked_inode_to_wb_end(struct inode *inode,
310 					    struct wb_lock_cookie *cookie)
311 {
312 	if (unlikely(cookie->locked))
313 		xa_unlock_irqrestore(&inode->i_mapping->i_pages, cookie->flags);
314 
315 	rcu_read_unlock();
316 }
317 
318 #else	/* CONFIG_CGROUP_WRITEBACK */
319 
320 static inline bool inode_cgwb_enabled(struct inode *inode)
321 {
322 	return false;
323 }
324 
325 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
326 {
327 	return &bdi->wb;
328 }
329 
330 static inline struct bdi_writeback *
331 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
332 {
333 	return &bdi->wb;
334 }
335 
336 static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
337 {
338 	return &inode_to_bdi(inode)->wb;
339 }
340 
341 static inline struct bdi_writeback *inode_to_wb_wbc(
342 				struct inode *inode,
343 				struct writeback_control *wbc)
344 {
345 	return inode_to_wb(inode);
346 }
347 
348 
349 static inline struct bdi_writeback *
350 unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie)
351 {
352 	return inode_to_wb(inode);
353 }
354 
355 static inline void unlocked_inode_to_wb_end(struct inode *inode,
356 					    struct wb_lock_cookie *cookie)
357 {
358 }
359 
360 static inline void wb_memcg_offline(struct mem_cgroup *memcg)
361 {
362 }
363 
364 static inline void wb_blkcg_offline(struct cgroup_subsys_state *css)
365 {
366 }
367 
368 #endif	/* CONFIG_CGROUP_WRITEBACK */
369 
370 const char *bdi_dev_name(struct backing_dev_info *bdi);
371 
372 #endif	/* _LINUX_BACKING_DEV_H */
373