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 
bdi_get(struct backing_dev_info * bdi)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 
wb_has_dirty_io(struct bdi_writeback * wb)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 
bdi_has_dirty_io(struct backing_dev_info * bdi)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 
wb_stat_mod(struct bdi_writeback * wb,enum wb_stat_item item,s64 amount)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 
inc_wb_stat(struct bdi_writeback * wb,enum wb_stat_item item)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 
dec_wb_stat(struct bdi_writeback * wb,enum wb_stat_item item)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 
wb_stat(struct bdi_writeback * wb,enum wb_stat_item item)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 
wb_stat_sum(struct bdi_writeback * wb,enum wb_stat_item item)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  */
wb_stat_error(void)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_min_ratio_no_scale(struct backing_dev_info *bdi, unsigned int min_ratio);
113 int bdi_set_max_ratio_no_scale(struct backing_dev_info *bdi, unsigned int max_ratio);
114 int bdi_set_min_bytes(struct backing_dev_info *bdi, u64 min_bytes);
115 int bdi_set_max_bytes(struct backing_dev_info *bdi, u64 max_bytes);
116 int bdi_set_strict_limit(struct backing_dev_info *bdi, unsigned int strict_limit);
117 
118 /*
119  * Flags in backing_dev_info::capability
120  *
121  * BDI_CAP_WRITEBACK:		Supports dirty page writeback, and dirty pages
122  *				should contribute to accounting
123  * BDI_CAP_WRITEBACK_ACCT:	Automatically account writeback pages
124  * BDI_CAP_STRICTLIMIT:		Keep number of dirty pages below bdi threshold
125  */
126 #define BDI_CAP_WRITEBACK		(1 << 0)
127 #define BDI_CAP_WRITEBACK_ACCT		(1 << 1)
128 #define BDI_CAP_STRICTLIMIT		(1 << 2)
129 
130 extern struct backing_dev_info noop_backing_dev_info;
131 
132 int bdi_init(struct backing_dev_info *bdi);
133 
134 /**
135  * writeback_in_progress - determine whether there is writeback in progress
136  * @wb: bdi_writeback of interest
137  *
138  * Determine whether there is writeback waiting to be handled against a
139  * bdi_writeback.
140  */
writeback_in_progress(struct bdi_writeback * wb)141 static inline bool writeback_in_progress(struct bdi_writeback *wb)
142 {
143 	return test_bit(WB_writeback_running, &wb->state);
144 }
145 
146 struct backing_dev_info *inode_to_bdi(struct inode *inode);
147 
mapping_can_writeback(struct address_space * mapping)148 static inline bool mapping_can_writeback(struct address_space *mapping)
149 {
150 	return inode_to_bdi(mapping->host)->capabilities & BDI_CAP_WRITEBACK;
151 }
152 
153 #ifdef CONFIG_CGROUP_WRITEBACK
154 
155 struct bdi_writeback *wb_get_lookup(struct backing_dev_info *bdi,
156 				    struct cgroup_subsys_state *memcg_css);
157 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
158 				    struct cgroup_subsys_state *memcg_css,
159 				    gfp_t gfp);
160 void wb_memcg_offline(struct mem_cgroup *memcg);
161 void wb_blkcg_offline(struct cgroup_subsys_state *css);
162 
163 /**
164  * inode_cgwb_enabled - test whether cgroup writeback is enabled on an inode
165  * @inode: inode of interest
166  *
167  * Cgroup writeback requires support from the filesystem.  Also, both memcg and
168  * iocg have to be on the default hierarchy.  Test whether all conditions are
169  * met.
170  *
171  * Note that the test result may change dynamically on the same inode
172  * depending on how memcg and iocg are configured.
173  */
inode_cgwb_enabled(struct inode * inode)174 static inline bool inode_cgwb_enabled(struct inode *inode)
175 {
176 	struct backing_dev_info *bdi = inode_to_bdi(inode);
177 
178 	return cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
179 		cgroup_subsys_on_dfl(io_cgrp_subsys) &&
180 		(bdi->capabilities & BDI_CAP_WRITEBACK) &&
181 		(inode->i_sb->s_iflags & SB_I_CGROUPWB);
182 }
183 
184 /**
185  * wb_find_current - find wb for %current on a bdi
186  * @bdi: bdi of interest
187  *
188  * Find the wb of @bdi which matches both the memcg and blkcg of %current.
189  * Must be called under rcu_read_lock() which protects the returend wb.
190  * NULL if not found.
191  */
wb_find_current(struct backing_dev_info * bdi)192 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
193 {
194 	struct cgroup_subsys_state *memcg_css;
195 	struct bdi_writeback *wb;
196 
197 	memcg_css = task_css(current, memory_cgrp_id);
198 	if (!memcg_css->parent)
199 		return &bdi->wb;
200 
201 	wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
202 
203 	/*
204 	 * %current's blkcg equals the effective blkcg of its memcg.  No
205 	 * need to use the relatively expensive cgroup_get_e_css().
206 	 */
207 	if (likely(wb && wb->blkcg_css == task_css(current, io_cgrp_id)))
208 		return wb;
209 	return NULL;
210 }
211 
212 /**
213  * wb_get_create_current - get or create wb for %current on a bdi
214  * @bdi: bdi of interest
215  * @gfp: allocation mask
216  *
217  * Equivalent to wb_get_create() on %current's memcg.  This function is
218  * called from a relatively hot path and optimizes the common cases using
219  * wb_find_current().
220  */
221 static inline struct bdi_writeback *
wb_get_create_current(struct backing_dev_info * bdi,gfp_t gfp)222 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
223 {
224 	struct bdi_writeback *wb;
225 
226 	rcu_read_lock();
227 	wb = wb_find_current(bdi);
228 	if (wb && unlikely(!wb_tryget(wb)))
229 		wb = NULL;
230 	rcu_read_unlock();
231 
232 	if (unlikely(!wb)) {
233 		struct cgroup_subsys_state *memcg_css;
234 
235 		memcg_css = task_get_css(current, memory_cgrp_id);
236 		wb = wb_get_create(bdi, memcg_css, gfp);
237 		css_put(memcg_css);
238 	}
239 	return wb;
240 }
241 
242 /**
243  * inode_to_wb - determine the wb of an inode
244  * @inode: inode of interest
245  *
246  * Returns the wb @inode is currently associated with.  The caller must be
247  * holding either @inode->i_lock, the i_pages lock, or the
248  * associated wb's list_lock.
249  */
inode_to_wb(const struct inode * inode)250 static inline struct bdi_writeback *inode_to_wb(const struct inode *inode)
251 {
252 #ifdef CONFIG_LOCKDEP
253 	WARN_ON_ONCE(debug_locks &&
254 		     (!lockdep_is_held(&inode->i_lock) &&
255 		      !lockdep_is_held(&inode->i_mapping->i_pages.xa_lock) &&
256 		      !lockdep_is_held(&inode->i_wb->list_lock)));
257 #endif
258 	return inode->i_wb;
259 }
260 
inode_to_wb_wbc(struct inode * inode,struct writeback_control * wbc)261 static inline struct bdi_writeback *inode_to_wb_wbc(
262 				struct inode *inode,
263 				struct writeback_control *wbc)
264 {
265 	/*
266 	 * If wbc does not have inode attached, it means cgroup writeback was
267 	 * disabled when wbc started. Just use the default wb in that case.
268 	 */
269 	return wbc->wb ? wbc->wb : &inode_to_bdi(inode)->wb;
270 }
271 
272 /**
273  * unlocked_inode_to_wb_begin - begin unlocked inode wb access transaction
274  * @inode: target inode
275  * @cookie: output param, to be passed to the end function
276  *
277  * The caller wants to access the wb associated with @inode but isn't
278  * holding inode->i_lock, the i_pages lock or wb->list_lock.  This
279  * function determines the wb associated with @inode and ensures that the
280  * association doesn't change until the transaction is finished with
281  * unlocked_inode_to_wb_end().
282  *
283  * The caller must call unlocked_inode_to_wb_end() with *@cookie afterwards and
284  * can't sleep during the transaction.  IRQs may or may not be disabled on
285  * return.
286  */
287 static inline struct bdi_writeback *
unlocked_inode_to_wb_begin(struct inode * inode,struct wb_lock_cookie * cookie)288 unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie)
289 {
290 	rcu_read_lock();
291 
292 	/*
293 	 * Paired with store_release in inode_switch_wbs_work_fn() and
294 	 * ensures that we see the new wb if we see cleared I_WB_SWITCH.
295 	 */
296 	cookie->locked = smp_load_acquire(&inode->i_state) & I_WB_SWITCH;
297 
298 	if (unlikely(cookie->locked))
299 		xa_lock_irqsave(&inode->i_mapping->i_pages, cookie->flags);
300 
301 	/*
302 	 * Protected by either !I_WB_SWITCH + rcu_read_lock() or the i_pages
303 	 * lock.  inode_to_wb() will bark.  Deref directly.
304 	 */
305 	return inode->i_wb;
306 }
307 
308 /**
309  * unlocked_inode_to_wb_end - end inode wb access transaction
310  * @inode: target inode
311  * @cookie: @cookie from unlocked_inode_to_wb_begin()
312  */
unlocked_inode_to_wb_end(struct inode * inode,struct wb_lock_cookie * cookie)313 static inline void unlocked_inode_to_wb_end(struct inode *inode,
314 					    struct wb_lock_cookie *cookie)
315 {
316 	if (unlikely(cookie->locked))
317 		xa_unlock_irqrestore(&inode->i_mapping->i_pages, cookie->flags);
318 
319 	rcu_read_unlock();
320 }
321 
322 #else	/* CONFIG_CGROUP_WRITEBACK */
323 
inode_cgwb_enabled(struct inode * inode)324 static inline bool inode_cgwb_enabled(struct inode *inode)
325 {
326 	return false;
327 }
328 
wb_find_current(struct backing_dev_info * bdi)329 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
330 {
331 	return &bdi->wb;
332 }
333 
334 static inline struct bdi_writeback *
wb_get_create_current(struct backing_dev_info * bdi,gfp_t gfp)335 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
336 {
337 	return &bdi->wb;
338 }
339 
inode_to_wb(struct inode * inode)340 static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
341 {
342 	return &inode_to_bdi(inode)->wb;
343 }
344 
inode_to_wb_wbc(struct inode * inode,struct writeback_control * wbc)345 static inline struct bdi_writeback *inode_to_wb_wbc(
346 				struct inode *inode,
347 				struct writeback_control *wbc)
348 {
349 	return inode_to_wb(inode);
350 }
351 
352 
353 static inline struct bdi_writeback *
unlocked_inode_to_wb_begin(struct inode * inode,struct wb_lock_cookie * cookie)354 unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie)
355 {
356 	return inode_to_wb(inode);
357 }
358 
unlocked_inode_to_wb_end(struct inode * inode,struct wb_lock_cookie * cookie)359 static inline void unlocked_inode_to_wb_end(struct inode *inode,
360 					    struct wb_lock_cookie *cookie)
361 {
362 }
363 
wb_memcg_offline(struct mem_cgroup * memcg)364 static inline void wb_memcg_offline(struct mem_cgroup *memcg)
365 {
366 }
367 
wb_blkcg_offline(struct cgroup_subsys_state * css)368 static inline void wb_blkcg_offline(struct cgroup_subsys_state *css)
369 {
370 }
371 
372 #endif	/* CONFIG_CGROUP_WRITEBACK */
373 
374 const char *bdi_dev_name(struct backing_dev_info *bdi);
375 
376 #endif	/* _LINUX_BACKING_DEV_H */
377