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/blkdev.h> |
16 | #include <linux/writeback.h> |
17 | #include <linux/blk-cgroup.h> |
18 | #include <linux/backing-dev-defs.h> |
19 | #include <linux/slab.h> |
20 | |
21 | static inline struct backing_dev_info *bdi_get(struct backing_dev_info *bdi) |
22 | { |
23 | kref_get(&bdi->refcnt); |
24 | return bdi; |
25 | } |
26 | |
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 | int bdi_register_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_node(gfp_t gfp_mask, int node_id); |
38 | static inline struct backing_dev_info *bdi_alloc(gfp_t gfp_mask) |
39 | { |
40 | return bdi_alloc_node(gfp_mask, NUMA_NO_NODE); |
41 | } |
42 | |
43 | void wb_start_background_writeback(struct bdi_writeback *wb); |
44 | void wb_workfn(struct work_struct *work); |
45 | void wb_wakeup_delayed(struct bdi_writeback *wb); |
46 | |
47 | extern spinlock_t bdi_lock; |
48 | extern struct list_head bdi_list; |
49 | |
50 | extern struct workqueue_struct *bdi_wq; |
51 | |
52 | static inline bool wb_has_dirty_io(struct bdi_writeback *wb) |
53 | { |
54 | return test_bit(WB_has_dirty_io, &wb->state); |
55 | } |
56 | |
57 | static inline bool bdi_has_dirty_io(struct backing_dev_info *bdi) |
58 | { |
59 | /* |
60 | * @bdi->tot_write_bandwidth is guaranteed to be > 0 if there are |
61 | * any dirty wbs. See wb_update_write_bandwidth(). |
62 | */ |
63 | return atomic_long_read(&bdi->tot_write_bandwidth); |
64 | } |
65 | |
66 | static inline void __add_wb_stat(struct bdi_writeback *wb, |
67 | enum wb_stat_item item, s64 amount) |
68 | { |
69 | percpu_counter_add_batch(&wb->stat[item], amount, WB_STAT_BATCH); |
70 | } |
71 | |
72 | static inline void inc_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item) |
73 | { |
74 | __add_wb_stat(wb, item, 1); |
75 | } |
76 | |
77 | static inline void dec_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item) |
78 | { |
79 | __add_wb_stat(wb, item, -1); |
80 | } |
81 | |
82 | static inline s64 wb_stat(struct bdi_writeback *wb, enum wb_stat_item item) |
83 | { |
84 | return percpu_counter_read_positive(&wb->stat[item]); |
85 | } |
86 | |
87 | static inline s64 wb_stat_sum(struct bdi_writeback *wb, enum wb_stat_item item) |
88 | { |
89 | return percpu_counter_sum_positive(&wb->stat[item]); |
90 | } |
91 | |
92 | extern void wb_writeout_inc(struct bdi_writeback *wb); |
93 | |
94 | /* |
95 | * maximal error of a stat counter. |
96 | */ |
97 | static inline unsigned long wb_stat_error(void) |
98 | { |
99 | #ifdef CONFIG_SMP |
100 | return nr_cpu_ids * WB_STAT_BATCH; |
101 | #else |
102 | return 1; |
103 | #endif |
104 | } |
105 | |
106 | int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio); |
107 | int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio); |
108 | |
109 | /* |
110 | * Flags in backing_dev_info::capability |
111 | * |
112 | * The first three flags control whether dirty pages will contribute to the |
113 | * VM's accounting and whether writepages() should be called for dirty pages |
114 | * (something that would not, for example, be appropriate for ramfs) |
115 | * |
116 | * WARNING: these flags are closely related and should not normally be |
117 | * used separately. The BDI_CAP_NO_ACCT_AND_WRITEBACK combines these |
118 | * three flags into a single convenience macro. |
119 | * |
120 | * BDI_CAP_NO_ACCT_DIRTY: Dirty pages shouldn't contribute to accounting |
121 | * BDI_CAP_NO_WRITEBACK: Don't write pages back |
122 | * BDI_CAP_NO_ACCT_WB: Don't automatically account writeback pages |
123 | * BDI_CAP_STRICTLIMIT: Keep number of dirty pages below bdi threshold. |
124 | * |
125 | * BDI_CAP_CGROUP_WRITEBACK: Supports cgroup-aware writeback. |
126 | * BDI_CAP_SYNCHRONOUS_IO: Device is so fast that asynchronous IO would be |
127 | * inefficient. |
128 | */ |
129 | #define BDI_CAP_NO_ACCT_DIRTY 0x00000001 |
130 | #define BDI_CAP_NO_WRITEBACK 0x00000002 |
131 | #define BDI_CAP_NO_ACCT_WB 0x00000004 |
132 | #define BDI_CAP_STABLE_WRITES 0x00000008 |
133 | #define BDI_CAP_STRICTLIMIT 0x00000010 |
134 | #define BDI_CAP_CGROUP_WRITEBACK 0x00000020 |
135 | #define BDI_CAP_SYNCHRONOUS_IO 0x00000040 |
136 | |
137 | #define BDI_CAP_NO_ACCT_AND_WRITEBACK \ |
138 | (BDI_CAP_NO_WRITEBACK | BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_ACCT_WB) |
139 | |
140 | extern struct backing_dev_info noop_backing_dev_info; |
141 | |
142 | /** |
143 | * writeback_in_progress - determine whether there is writeback in progress |
144 | * @wb: bdi_writeback of interest |
145 | * |
146 | * Determine whether there is writeback waiting to be handled against a |
147 | * bdi_writeback. |
148 | */ |
149 | static inline bool writeback_in_progress(struct bdi_writeback *wb) |
150 | { |
151 | return test_bit(WB_writeback_running, &wb->state); |
152 | } |
153 | |
154 | static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) |
155 | { |
156 | struct super_block *sb; |
157 | |
158 | if (!inode) |
159 | return &noop_backing_dev_info; |
160 | |
161 | sb = inode->i_sb; |
162 | #ifdef CONFIG_BLOCK |
163 | if (sb_is_blkdev_sb(sb)) |
164 | return I_BDEV(inode)->bd_bdi; |
165 | #endif |
166 | return sb->s_bdi; |
167 | } |
168 | |
169 | static inline int wb_congested(struct bdi_writeback *wb, int cong_bits) |
170 | { |
171 | struct backing_dev_info *bdi = wb->bdi; |
172 | |
173 | if (bdi->congested_fn) |
174 | return bdi->congested_fn(bdi->congested_data, cong_bits); |
175 | return wb->congested->state & cong_bits; |
176 | } |
177 | |
178 | long congestion_wait(int sync, long timeout); |
179 | long wait_iff_congested(int sync, long timeout); |
180 | |
181 | static inline bool bdi_cap_synchronous_io(struct backing_dev_info *bdi) |
182 | { |
183 | return bdi->capabilities & BDI_CAP_SYNCHRONOUS_IO; |
184 | } |
185 | |
186 | static inline bool bdi_cap_stable_pages_required(struct backing_dev_info *bdi) |
187 | { |
188 | return bdi->capabilities & BDI_CAP_STABLE_WRITES; |
189 | } |
190 | |
191 | static inline bool bdi_cap_writeback_dirty(struct backing_dev_info *bdi) |
192 | { |
193 | return !(bdi->capabilities & BDI_CAP_NO_WRITEBACK); |
194 | } |
195 | |
196 | static inline bool bdi_cap_account_dirty(struct backing_dev_info *bdi) |
197 | { |
198 | return !(bdi->capabilities & BDI_CAP_NO_ACCT_DIRTY); |
199 | } |
200 | |
201 | static inline bool bdi_cap_account_writeback(struct backing_dev_info *bdi) |
202 | { |
203 | /* Paranoia: BDI_CAP_NO_WRITEBACK implies BDI_CAP_NO_ACCT_WB */ |
204 | return !(bdi->capabilities & (BDI_CAP_NO_ACCT_WB | |
205 | BDI_CAP_NO_WRITEBACK)); |
206 | } |
207 | |
208 | static inline bool mapping_cap_writeback_dirty(struct address_space *mapping) |
209 | { |
210 | return bdi_cap_writeback_dirty(inode_to_bdi(mapping->host)); |
211 | } |
212 | |
213 | static inline bool mapping_cap_account_dirty(struct address_space *mapping) |
214 | { |
215 | return bdi_cap_account_dirty(inode_to_bdi(mapping->host)); |
216 | } |
217 | |
218 | static inline int bdi_sched_wait(void *word) |
219 | { |
220 | schedule(); |
221 | return 0; |
222 | } |
223 | |
224 | #ifdef CONFIG_CGROUP_WRITEBACK |
225 | |
226 | struct bdi_writeback_congested * |
227 | wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp); |
228 | void wb_congested_put(struct bdi_writeback_congested *congested); |
229 | struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi, |
230 | struct cgroup_subsys_state *memcg_css, |
231 | gfp_t gfp); |
232 | void wb_memcg_offline(struct mem_cgroup *memcg); |
233 | void wb_blkcg_offline(struct blkcg *blkcg); |
234 | int inode_congested(struct inode *inode, int cong_bits); |
235 | |
236 | /** |
237 | * inode_cgwb_enabled - test whether cgroup writeback is enabled on an inode |
238 | * @inode: inode of interest |
239 | * |
240 | * cgroup writeback requires support from both the bdi and filesystem. |
241 | * Also, both memcg and iocg have to be on the default hierarchy. Test |
242 | * whether all conditions are met. |
243 | * |
244 | * Note that the test result may change dynamically on the same inode |
245 | * depending on how memcg and iocg are configured. |
246 | */ |
247 | static inline bool inode_cgwb_enabled(struct inode *inode) |
248 | { |
249 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
250 | |
251 | return cgroup_subsys_on_dfl(memory_cgrp_subsys) && |
252 | cgroup_subsys_on_dfl(io_cgrp_subsys) && |
253 | bdi_cap_account_dirty(bdi) && |
254 | (bdi->capabilities & BDI_CAP_CGROUP_WRITEBACK) && |
255 | (inode->i_sb->s_iflags & SB_I_CGROUPWB); |
256 | } |
257 | |
258 | /** |
259 | * wb_find_current - find wb for %current on a bdi |
260 | * @bdi: bdi of interest |
261 | * |
262 | * Find the wb of @bdi which matches both the memcg and blkcg of %current. |
263 | * Must be called under rcu_read_lock() which protects the returend wb. |
264 | * NULL if not found. |
265 | */ |
266 | static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi) |
267 | { |
268 | struct cgroup_subsys_state *memcg_css; |
269 | struct bdi_writeback *wb; |
270 | |
271 | memcg_css = task_css(current, memory_cgrp_id); |
272 | if (!memcg_css->parent) |
273 | return &bdi->wb; |
274 | |
275 | wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id); |
276 | |
277 | /* |
278 | * %current's blkcg equals the effective blkcg of its memcg. No |
279 | * need to use the relatively expensive cgroup_get_e_css(). |
280 | */ |
281 | if (likely(wb && wb->blkcg_css == task_css(current, io_cgrp_id))) |
282 | return wb; |
283 | return NULL; |
284 | } |
285 | |
286 | /** |
287 | * wb_get_create_current - get or create wb for %current on a bdi |
288 | * @bdi: bdi of interest |
289 | * @gfp: allocation mask |
290 | * |
291 | * Equivalent to wb_get_create() on %current's memcg. This function is |
292 | * called from a relatively hot path and optimizes the common cases using |
293 | * wb_find_current(). |
294 | */ |
295 | static inline struct bdi_writeback * |
296 | wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp) |
297 | { |
298 | struct bdi_writeback *wb; |
299 | |
300 | rcu_read_lock(); |
301 | wb = wb_find_current(bdi); |
302 | if (wb && unlikely(!wb_tryget(wb))) |
303 | wb = NULL; |
304 | rcu_read_unlock(); |
305 | |
306 | if (unlikely(!wb)) { |
307 | struct cgroup_subsys_state *memcg_css; |
308 | |
309 | memcg_css = task_get_css(current, memory_cgrp_id); |
310 | wb = wb_get_create(bdi, memcg_css, gfp); |
311 | css_put(memcg_css); |
312 | } |
313 | return wb; |
314 | } |
315 | |
316 | /** |
317 | * inode_to_wb_is_valid - test whether an inode has a wb associated |
318 | * @inode: inode of interest |
319 | * |
320 | * Returns %true if @inode has a wb associated. May be called without any |
321 | * locking. |
322 | */ |
323 | static inline bool inode_to_wb_is_valid(struct inode *inode) |
324 | { |
325 | return inode->i_wb; |
326 | } |
327 | |
328 | /** |
329 | * inode_to_wb - determine the wb of an inode |
330 | * @inode: inode of interest |
331 | * |
332 | * Returns the wb @inode is currently associated with. The caller must be |
333 | * holding either @inode->i_lock, the i_pages lock, or the |
334 | * associated wb's list_lock. |
335 | */ |
336 | static inline struct bdi_writeback *inode_to_wb(const struct inode *inode) |
337 | { |
338 | #ifdef CONFIG_LOCKDEP |
339 | WARN_ON_ONCE(debug_locks && |
340 | (!lockdep_is_held(&inode->i_lock) && |
341 | !lockdep_is_held(&inode->i_mapping->i_pages.xa_lock) && |
342 | !lockdep_is_held(&inode->i_wb->list_lock))); |
343 | #endif |
344 | return inode->i_wb; |
345 | } |
346 | |
347 | /** |
348 | * unlocked_inode_to_wb_begin - begin unlocked inode wb access transaction |
349 | * @inode: target inode |
350 | * @cookie: output param, to be passed to the end function |
351 | * |
352 | * The caller wants to access the wb associated with @inode but isn't |
353 | * holding inode->i_lock, the i_pages lock or wb->list_lock. This |
354 | * function determines the wb associated with @inode and ensures that the |
355 | * association doesn't change until the transaction is finished with |
356 | * unlocked_inode_to_wb_end(). |
357 | * |
358 | * The caller must call unlocked_inode_to_wb_end() with *@cookie afterwards and |
359 | * can't sleep during the transaction. IRQs may or may not be disabled on |
360 | * return. |
361 | */ |
362 | static inline struct bdi_writeback * |
363 | unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie) |
364 | { |
365 | rcu_read_lock(); |
366 | |
367 | /* |
368 | * Paired with store_release in inode_switch_wbs_work_fn() and |
369 | * ensures that we see the new wb if we see cleared I_WB_SWITCH. |
370 | */ |
371 | cookie->locked = smp_load_acquire(&inode->i_state) & I_WB_SWITCH; |
372 | |
373 | if (unlikely(cookie->locked)) |
374 | xa_lock_irqsave(&inode->i_mapping->i_pages, cookie->flags); |
375 | |
376 | /* |
377 | * Protected by either !I_WB_SWITCH + rcu_read_lock() or the i_pages |
378 | * lock. inode_to_wb() will bark. Deref directly. |
379 | */ |
380 | return inode->i_wb; |
381 | } |
382 | |
383 | /** |
384 | * unlocked_inode_to_wb_end - end inode wb access transaction |
385 | * @inode: target inode |
386 | * @cookie: @cookie from unlocked_inode_to_wb_begin() |
387 | */ |
388 | static inline void unlocked_inode_to_wb_end(struct inode *inode, |
389 | struct wb_lock_cookie *cookie) |
390 | { |
391 | if (unlikely(cookie->locked)) |
392 | xa_unlock_irqrestore(&inode->i_mapping->i_pages, cookie->flags); |
393 | |
394 | rcu_read_unlock(); |
395 | } |
396 | |
397 | #else /* CONFIG_CGROUP_WRITEBACK */ |
398 | |
399 | static inline bool inode_cgwb_enabled(struct inode *inode) |
400 | { |
401 | return false; |
402 | } |
403 | |
404 | static inline struct bdi_writeback_congested * |
405 | wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp) |
406 | { |
407 | refcount_inc(&bdi->wb_congested->refcnt); |
408 | return bdi->wb_congested; |
409 | } |
410 | |
411 | static inline void wb_congested_put(struct bdi_writeback_congested *congested) |
412 | { |
413 | if (refcount_dec_and_test(&congested->refcnt)) |
414 | kfree(congested); |
415 | } |
416 | |
417 | static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi) |
418 | { |
419 | return &bdi->wb; |
420 | } |
421 | |
422 | static inline struct bdi_writeback * |
423 | wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp) |
424 | { |
425 | return &bdi->wb; |
426 | } |
427 | |
428 | static inline bool inode_to_wb_is_valid(struct inode *inode) |
429 | { |
430 | return true; |
431 | } |
432 | |
433 | static inline struct bdi_writeback *inode_to_wb(struct inode *inode) |
434 | { |
435 | return &inode_to_bdi(inode)->wb; |
436 | } |
437 | |
438 | static inline struct bdi_writeback * |
439 | unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie) |
440 | { |
441 | return inode_to_wb(inode); |
442 | } |
443 | |
444 | static inline void unlocked_inode_to_wb_end(struct inode *inode, |
445 | struct wb_lock_cookie *cookie) |
446 | { |
447 | } |
448 | |
449 | static inline void wb_memcg_offline(struct mem_cgroup *memcg) |
450 | { |
451 | } |
452 | |
453 | static inline void wb_blkcg_offline(struct blkcg *blkcg) |
454 | { |
455 | } |
456 | |
457 | static inline int inode_congested(struct inode *inode, int cong_bits) |
458 | { |
459 | return wb_congested(&inode_to_bdi(inode)->wb, cong_bits); |
460 | } |
461 | |
462 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
463 | |
464 | static inline int inode_read_congested(struct inode *inode) |
465 | { |
466 | return inode_congested(inode, 1 << WB_sync_congested); |
467 | } |
468 | |
469 | static inline int inode_write_congested(struct inode *inode) |
470 | { |
471 | return inode_congested(inode, 1 << WB_async_congested); |
472 | } |
473 | |
474 | static inline int inode_rw_congested(struct inode *inode) |
475 | { |
476 | return inode_congested(inode, (1 << WB_sync_congested) | |
477 | (1 << WB_async_congested)); |
478 | } |
479 | |
480 | static inline int bdi_congested(struct backing_dev_info *bdi, int cong_bits) |
481 | { |
482 | return wb_congested(&bdi->wb, cong_bits); |
483 | } |
484 | |
485 | static inline int bdi_read_congested(struct backing_dev_info *bdi) |
486 | { |
487 | return bdi_congested(bdi, 1 << WB_sync_congested); |
488 | } |
489 | |
490 | static inline int bdi_write_congested(struct backing_dev_info *bdi) |
491 | { |
492 | return bdi_congested(bdi, 1 << WB_async_congested); |
493 | } |
494 | |
495 | static inline int bdi_rw_congested(struct backing_dev_info *bdi) |
496 | { |
497 | return bdi_congested(bdi, (1 << WB_sync_congested) | |
498 | (1 << WB_async_congested)); |
499 | } |
500 | |
501 | #endif /* _LINUX_BACKING_DEV_H */ |
502 | |