projects/linux·blob master · refs

375 lines · 10.4 KiB

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
20static inline struct backing_dev_info *bdi_get(struct backing_dev_info *bdi)
21{
22 kref_get(&bdi->refcnt);
23 return bdi;
24}
25
26struct backing_dev_info *bdi_get_by_id(u64 id);
27void bdi_put(struct backing_dev_info *bdi);
28
29__printf(2, 3)
30int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...);
31__printf(2, 0)
32int bdi_register_va(struct backing_dev_info *bdi, const char *fmt,
33 va_list args);
34void bdi_set_owner(struct backing_dev_info *bdi, struct device *owner);
35void bdi_unregister(struct backing_dev_info *bdi);
36
37struct backing_dev_info *bdi_alloc(int node_id);
38
39void wb_start_background_writeback(struct bdi_writeback *wb);
40void wb_workfn(struct work_struct *work);
41
42void wb_wait_for_completion(struct wb_completion *done);
43
44extern spinlock_t bdi_lock;
45extern struct list_head bdi_list;
46
47extern struct workqueue_struct *bdi_wq;
48
49static inline bool wb_has_dirty_io(struct bdi_writeback *wb)
50{
51 return test_bit(WB_has_dirty_io, &wb->state);
52}
53
54static inline bool bdi_has_dirty_io(struct backing_dev_info *bdi)
55{
56 /*
57 * @bdi->tot_write_bandwidth is guaranteed to be > 0 if there are
58 * any dirty wbs. See wb_update_write_bandwidth().
59 */
60 return atomic_long_read(&bdi->tot_write_bandwidth);
61}
62
63static inline void wb_stat_mod(struct bdi_writeback *wb,
64 enum wb_stat_item item, s64 amount)
65{
66 percpu_counter_add_batch(&wb->stat[item], amount, WB_STAT_BATCH);
67}
68
69static inline s64 wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
70{
71 return percpu_counter_read_positive(&wb->stat[item]);
72}
73
74static inline s64 wb_stat_sum(struct bdi_writeback *wb, enum wb_stat_item item)
75{
76 return percpu_counter_sum_positive(&wb->stat[item]);
77}
78
79/*
80 * maximal error of a stat counter.
81 */
82static inline unsigned long wb_stat_error(void)
83{
84#ifdef CONFIG_SMP
85 return nr_cpu_ids * WB_STAT_BATCH;
86#else
87 return 1;
88#endif
89}
90
91/* BDI ratio is expressed as part per 1000000 for finer granularity. */
92#define BDI_RATIO_SCALE 10000
93
94u64 bdi_get_min_bytes(struct backing_dev_info *bdi);
95u64 bdi_get_max_bytes(struct backing_dev_info *bdi);
96int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
97int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
98int bdi_set_min_ratio_no_scale(struct backing_dev_info *bdi, unsigned int min_ratio);
99int bdi_set_max_ratio_no_scale(struct backing_dev_info *bdi, unsigned int max_ratio);
100int bdi_set_min_bytes(struct backing_dev_info *bdi, u64 min_bytes);
101int bdi_set_max_bytes(struct backing_dev_info *bdi, u64 max_bytes);
102int bdi_set_strict_limit(struct backing_dev_info *bdi, unsigned int strict_limit);
103
104/*
105 * Flags in backing_dev_info::capability
106 *
107 * BDI_CAP_WRITEBACK: Supports dirty page writeback, and dirty pages
108 * should contribute to accounting
109 * BDI_CAP_STRICTLIMIT: Keep number of dirty pages below bdi threshold
110 */
111#define BDI_CAP_WRITEBACK (1 << 0)
112#define BDI_CAP_STRICTLIMIT (1 << 1)
113
114extern struct backing_dev_info noop_backing_dev_info;
115
116int bdi_init(struct backing_dev_info *bdi);
117
118/**
119 * writeback_in_progress - determine whether there is writeback in progress
120 * @wb: bdi_writeback of interest
121 *
122 * Determine whether there is writeback waiting to be handled against a
123 * bdi_writeback.
124 */
125static inline bool writeback_in_progress(struct bdi_writeback *wb)
126{
127 return test_bit(WB_writeback_running, &wb->state);
128}
129
130struct backing_dev_info *inode_to_bdi(struct inode *inode);
131
132static inline bool mapping_can_writeback(struct address_space *mapping)
133{
134 return inode_to_bdi(mapping->host)->capabilities & BDI_CAP_WRITEBACK;
135}
136
137/* Must not be used by file systems that support cgroup writeback */
138static inline int bdi_wb_dirty_exceeded(struct backing_dev_info *bdi)
139{
140 return bdi->wb.dirty_exceeded;
141}
142
143/* Must not be used by file systems that support cgroup writeback */
144static inline void bdi_wb_stat_mod(struct inode *inode, enum wb_stat_item item,
145 s64 amount)
146{
147 wb_stat_mod(&inode_to_bdi(inode)->wb, item, amount);
148}
149
150#ifdef CONFIG_CGROUP_WRITEBACK
151
152struct bdi_writeback *wb_get_lookup(struct backing_dev_info *bdi,
153 struct cgroup_subsys_state *memcg_css);
154struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
155 struct cgroup_subsys_state *memcg_css,
156 gfp_t gfp);
157void wb_memcg_offline(struct mem_cgroup *memcg);
158void 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 */
171static 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 */
189static 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 */
218static inline struct bdi_writeback *
219wb_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 */
247static inline struct bdi_writeback *inode_to_wb(const struct inode *inode)
248{
249#ifdef CONFIG_LOCKDEP
250 WARN_ON_ONCE(debug_locks &&
251 (inode->i_sb->s_iflags & SB_I_CGROUPWB) &&
252 (!lockdep_is_held(&inode->i_lock) &&
253 !lockdep_is_held(&inode->i_mapping->i_pages.xa_lock) &&
254 !lockdep_is_held(&inode->i_wb->list_lock)));
255#endif
256 return inode->i_wb;
257}
258
259static inline struct bdi_writeback *inode_to_wb_wbc(
260 struct inode *inode,
261 struct writeback_control *wbc)
262{
263 /*
264 * If wbc does not have inode attached, it means cgroup writeback was
265 * disabled when wbc started. Just use the default wb in that case.
266 */
267 return wbc->wb ? wbc->wb : &inode_to_bdi(inode)->wb;
268}
269
270/**
271 * unlocked_inode_to_wb_begin - begin unlocked inode wb access transaction
272 * @inode: target inode
273 * @cookie: output param, to be passed to the end function
274 *
275 * The caller wants to access the wb associated with @inode but isn't
276 * holding inode->i_lock, the i_pages lock or wb->list_lock. This
277 * function determines the wb associated with @inode and ensures that the
278 * association doesn't change until the transaction is finished with
279 * unlocked_inode_to_wb_end().
280 *
281 * The caller must call unlocked_inode_to_wb_end() with *@cookie afterwards and
282 * can't sleep during the transaction. IRQs may or may not be disabled on
283 * return.
284 */
285static inline struct bdi_writeback *
286unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie)
287{
288 rcu_read_lock();
289
290 /*
291 * Paired with a release fence in inode_do_switch_wbs() and
292 * ensures that we see the new wb if we see cleared I_WB_SWITCH.
293 */
294 cookie->locked = inode_state_read_once(inode) & I_WB_SWITCH;
295 smp_rmb();
296
297 if (unlikely(cookie->locked))
298 xa_lock_irqsave(&inode->i_mapping->i_pages, cookie->flags);
299
300 /*
301 * Protected by either !I_WB_SWITCH + rcu_read_lock() or the i_pages
302 * lock. inode_to_wb() will bark. Deref directly.
303 */
304 return inode->i_wb;
305}
306
307/**
308 * unlocked_inode_to_wb_end - end inode wb access transaction
309 * @inode: target inode
310 * @cookie: @cookie from unlocked_inode_to_wb_begin()
311 */
312static inline void unlocked_inode_to_wb_end(struct inode *inode,
313 struct wb_lock_cookie *cookie)
314{
315 if (unlikely(cookie->locked))
316 xa_unlock_irqrestore(&inode->i_mapping->i_pages, cookie->flags);
317
318 rcu_read_unlock();
319}
320
321#else /* CONFIG_CGROUP_WRITEBACK */
322
323static inline bool inode_cgwb_enabled(struct inode *inode)
324{
325 return false;
326}
327
328static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
329{
330 return &bdi->wb;
331}
332
333static inline struct bdi_writeback *
334wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
335{
336 return &bdi->wb;
337}
338
339static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
340{
341 return &inode_to_bdi(inode)->wb;
342}
343
344static inline struct bdi_writeback *inode_to_wb_wbc(
345 struct inode *inode,
346 struct writeback_control *wbc)
347{
348 return inode_to_wb(inode);
349}
350
351
352static inline struct bdi_writeback *
353unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie)
354{
355 return inode_to_wb(inode);
356}
357
358static inline void unlocked_inode_to_wb_end(struct inode *inode,
359 struct wb_lock_cookie *cookie)
360{
361}
362
363static inline void wb_memcg_offline(struct mem_cgroup *memcg)
364{
365}
366
367static inline void wb_blkcg_offline(struct cgroup_subsys_state *css)
368{
369}
370
371#endif /* CONFIG_CGROUP_WRITEBACK */
372
373const char *bdi_dev_name(struct backing_dev_info *bdi);
374
375#endif /* _LINUX_BACKING_DEV_H */