1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3 * include/linux/writeback.h
4 */
5 #ifndef WRITEBACK_H
6 #define WRITEBACK_H
7
8 #include <linux/sched.h>
9 #include <linux/workqueue.h>
10 #include <linux/fs.h>
11 #include <linux/flex_proportions.h>
12 #include <linux/backing-dev-defs.h>
13 #include <linux/blk_types.h>
14 #include <linux/pagevec.h>
15
16 struct bio;
17
18 DECLARE_PER_CPU(int, dirty_throttle_leaks);
19
20 /*
21 * The global dirty threshold is normally equal to the global dirty limit,
22 * except when the system suddenly allocates a lot of anonymous memory and
23 * knocks down the global dirty threshold quickly, in which case the global
24 * dirty limit will follow down slowly to prevent livelocking all dirtier tasks.
25 */
26 #define DIRTY_SCOPE 8
27
28 struct backing_dev_info;
29
30 /*
31 * fs/fs-writeback.c
32 */
33 enum writeback_sync_modes {
34 WB_SYNC_NONE, /* Don't wait on anything */
35 WB_SYNC_ALL, /* Wait on every mapping */
36 };
37
38 /*
39 * A control structure which tells the writeback code what to do. These are
40 * always on the stack, and hence need no locking. They are always initialised
41 * in a manner such that unspecified fields are set to zero.
42 */
43 struct writeback_control {
44 /* public fields that can be set and/or consumed by the caller: */
45 long nr_to_write; /* Write this many pages, and decrement
46 this for each page written */
47 long pages_skipped; /* Pages which were not written */
48
49 /*
50 * For a_ops->writepages(): if start or end are non-zero then this is
51 * a hint that the filesystem need only write out the pages inside that
52 * byterange. The byte at `end' is included in the writeout request.
53 */
54 loff_t range_start;
55 loff_t range_end;
56
57 enum writeback_sync_modes sync_mode;
58
59 unsigned for_kupdate:1; /* A kupdate writeback */
60 unsigned for_background:1; /* A background writeback */
61 unsigned tagged_writepages:1; /* tag-and-write to avoid livelock */
62 unsigned for_reclaim:1; /* Invoked from the page allocator */
63 unsigned range_cyclic:1; /* range_start is cyclic */
64 unsigned for_sync:1; /* sync(2) WB_SYNC_ALL writeback */
65 unsigned unpinned_netfs_wb:1; /* Cleared I_PINNING_NETFS_WB */
66
67 /*
68 * When writeback IOs are bounced through async layers, only the
69 * initial synchronous phase should be accounted towards inode
70 * cgroup ownership arbitration to avoid confusion. Later stages
71 * can set the following flag to disable the accounting.
72 */
73 unsigned no_cgroup_owner:1;
74
75 /* To enable batching of swap writes to non-block-device backends,
76 * "plug" can be set point to a 'struct swap_iocb *'. When all swap
77 * writes have been submitted, if with swap_iocb is not NULL,
78 * swap_write_unplug() should be called.
79 */
80 struct swap_iocb **swap_plug;
81
82 /* Target list for splitting a large folio */
83 struct list_head *list;
84
85 /* internal fields used by the ->writepages implementation: */
86 struct folio_batch fbatch;
87 pgoff_t index;
88 int saved_err;
89
90 #ifdef CONFIG_CGROUP_WRITEBACK
91 struct bdi_writeback *wb; /* wb this writeback is issued under */
92 struct inode *inode; /* inode being written out */
93
94 /* foreign inode detection, see wbc_detach_inode() */
95 int wb_id; /* current wb id */
96 int wb_lcand_id; /* last foreign candidate wb id */
97 int wb_tcand_id; /* this foreign candidate wb id */
98 size_t wb_bytes; /* bytes written by current wb */
99 size_t wb_lcand_bytes; /* bytes written by last candidate */
100 size_t wb_tcand_bytes; /* bytes written by this candidate */
101 #endif
102 };
103
wbc_to_write_flags(struct writeback_control * wbc)104 static inline blk_opf_t wbc_to_write_flags(struct writeback_control *wbc)
105 {
106 blk_opf_t flags = 0;
107
108 if (wbc->sync_mode == WB_SYNC_ALL)
109 flags |= REQ_SYNC;
110 else if (wbc->for_kupdate || wbc->for_background)
111 flags |= REQ_BACKGROUND;
112
113 return flags;
114 }
115
116 #ifdef CONFIG_CGROUP_WRITEBACK
117 #define wbc_blkcg_css(wbc) \
118 ((wbc)->wb ? (wbc)->wb->blkcg_css : blkcg_root_css)
119 #else
120 #define wbc_blkcg_css(wbc) (blkcg_root_css)
121 #endif /* CONFIG_CGROUP_WRITEBACK */
122
123 /*
124 * A wb_domain represents a domain that wb's (bdi_writeback's) belong to
125 * and are measured against each other in. There always is one global
126 * domain, global_wb_domain, that every wb in the system is a member of.
127 * This allows measuring the relative bandwidth of each wb to distribute
128 * dirtyable memory accordingly.
129 */
130 struct wb_domain {
131 spinlock_t lock;
132
133 /*
134 * Scale the writeback cache size proportional to the relative
135 * writeout speed.
136 *
137 * We do this by keeping a floating proportion between BDIs, based
138 * on page writeback completions [end_page_writeback()]. Those
139 * devices that write out pages fastest will get the larger share,
140 * while the slower will get a smaller share.
141 *
142 * We use page writeout completions because we are interested in
143 * getting rid of dirty pages. Having them written out is the
144 * primary goal.
145 *
146 * We introduce a concept of time, a period over which we measure
147 * these events, because demand can/will vary over time. The length
148 * of this period itself is measured in page writeback completions.
149 */
150 struct fprop_global completions;
151 struct timer_list period_timer; /* timer for aging of completions */
152 unsigned long period_time;
153
154 /*
155 * The dirtyable memory and dirty threshold could be suddenly
156 * knocked down by a large amount (eg. on the startup of KVM in a
157 * swapless system). This may throw the system into deep dirty
158 * exceeded state and throttle heavy/light dirtiers alike. To
159 * retain good responsiveness, maintain global_dirty_limit for
160 * tracking slowly down to the knocked down dirty threshold.
161 *
162 * Both fields are protected by ->lock.
163 */
164 unsigned long dirty_limit_tstamp;
165 unsigned long dirty_limit;
166 };
167
168 /**
169 * wb_domain_size_changed - memory available to a wb_domain has changed
170 * @dom: wb_domain of interest
171 *
172 * This function should be called when the amount of memory available to
173 * @dom has changed. It resets @dom's dirty limit parameters to prevent
174 * the past values which don't match the current configuration from skewing
175 * dirty throttling. Without this, when memory size of a wb_domain is
176 * greatly reduced, the dirty throttling logic may allow too many pages to
177 * be dirtied leading to consecutive unnecessary OOMs and may get stuck in
178 * that situation.
179 */
wb_domain_size_changed(struct wb_domain * dom)180 static inline void wb_domain_size_changed(struct wb_domain *dom)
181 {
182 spin_lock(&dom->lock);
183 dom->dirty_limit_tstamp = jiffies;
184 dom->dirty_limit = 0;
185 spin_unlock(&dom->lock);
186 }
187
188 /*
189 * fs/fs-writeback.c
190 */
191 struct bdi_writeback;
192 void writeback_inodes_sb(struct super_block *, enum wb_reason reason);
193 void writeback_inodes_sb_nr(struct super_block *, unsigned long nr,
194 enum wb_reason reason);
195 void try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason);
196 void sync_inodes_sb(struct super_block *);
197 void wakeup_flusher_threads(enum wb_reason reason);
198 void wakeup_flusher_threads_bdi(struct backing_dev_info *bdi,
199 enum wb_reason reason);
200 void inode_wait_for_writeback(struct inode *inode);
201 void inode_io_list_del(struct inode *inode);
202
203 /* writeback.h requires fs.h; it, too, is not included from here. */
wait_on_inode(struct inode * inode)204 static inline void wait_on_inode(struct inode *inode)
205 {
206 wait_var_event(inode_state_wait_address(inode, __I_NEW),
207 !(READ_ONCE(inode->i_state) & I_NEW));
208 }
209
210 #ifdef CONFIG_CGROUP_WRITEBACK
211
212 #include <linux/cgroup.h>
213 #include <linux/bio.h>
214
215 void __inode_attach_wb(struct inode *inode, struct folio *folio);
216 void wbc_detach_inode(struct writeback_control *wbc);
217 void wbc_account_cgroup_owner(struct writeback_control *wbc, struct folio *folio,
218 size_t bytes);
219 int cgroup_writeback_by_id(u64 bdi_id, int memcg_id,
220 enum wb_reason reason, struct wb_completion *done);
221 void cgroup_writeback_umount(struct super_block *sb);
222 bool cleanup_offline_cgwb(struct bdi_writeback *wb);
223
224 /**
225 * inode_attach_wb - associate an inode with its wb
226 * @inode: inode of interest
227 * @folio: folio being dirtied (may be NULL)
228 *
229 * If @inode doesn't have its wb, associate it with the wb matching the
230 * memcg of @folio or, if @folio is NULL, %current. May be called w/ or w/o
231 * @inode->i_lock.
232 */
inode_attach_wb(struct inode * inode,struct folio * folio)233 static inline void inode_attach_wb(struct inode *inode, struct folio *folio)
234 {
235 if (!inode->i_wb)
236 __inode_attach_wb(inode, folio);
237 }
238
239 /**
240 * inode_detach_wb - disassociate an inode from its wb
241 * @inode: inode of interest
242 *
243 * @inode is being freed. Detach from its wb.
244 */
inode_detach_wb(struct inode * inode)245 static inline void inode_detach_wb(struct inode *inode)
246 {
247 if (inode->i_wb) {
248 WARN_ON_ONCE(!(inode->i_state & I_CLEAR));
249 wb_put(inode->i_wb);
250 inode->i_wb = NULL;
251 }
252 }
253
254 void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
255 struct inode *inode);
256
257 /**
258 * wbc_init_bio - writeback specific initializtion of bio
259 * @wbc: writeback_control for the writeback in progress
260 * @bio: bio to be initialized
261 *
262 * @bio is a part of the writeback in progress controlled by @wbc. Perform
263 * writeback specific initialization. This is used to apply the cgroup
264 * writeback context. Must be called after the bio has been associated with
265 * a device.
266 */
wbc_init_bio(struct writeback_control * wbc,struct bio * bio)267 static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
268 {
269 /*
270 * pageout() path doesn't attach @wbc to the inode being written
271 * out. This is intentional as we don't want the function to block
272 * behind a slow cgroup. Ultimately, we want pageout() to kick off
273 * regular writeback instead of writing things out itself.
274 */
275 if (wbc->wb)
276 bio_associate_blkg_from_css(bio, wbc->wb->blkcg_css);
277 }
278
279 #else /* CONFIG_CGROUP_WRITEBACK */
280
inode_attach_wb(struct inode * inode,struct folio * folio)281 static inline void inode_attach_wb(struct inode *inode, struct folio *folio)
282 {
283 }
284
inode_detach_wb(struct inode * inode)285 static inline void inode_detach_wb(struct inode *inode)
286 {
287 }
288
wbc_attach_fdatawrite_inode(struct writeback_control * wbc,struct inode * inode)289 static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
290 struct inode *inode)
291 {
292 }
293
wbc_detach_inode(struct writeback_control * wbc)294 static inline void wbc_detach_inode(struct writeback_control *wbc)
295 {
296 }
297
wbc_init_bio(struct writeback_control * wbc,struct bio * bio)298 static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
299 {
300 }
301
wbc_account_cgroup_owner(struct writeback_control * wbc,struct folio * folio,size_t bytes)302 static inline void wbc_account_cgroup_owner(struct writeback_control *wbc,
303 struct folio *folio, size_t bytes)
304 {
305 }
306
cgroup_writeback_umount(struct super_block * sb)307 static inline void cgroup_writeback_umount(struct super_block *sb)
308 {
309 }
310
311 #endif /* CONFIG_CGROUP_WRITEBACK */
312
313 /*
314 * mm/page-writeback.c
315 */
316 void laptop_io_completion(struct backing_dev_info *info);
317 void laptop_sync_completion(void);
318 void laptop_mode_timer_fn(struct timer_list *t);
319 bool node_dirty_ok(struct pglist_data *pgdat);
320 int wb_domain_init(struct wb_domain *dom, gfp_t gfp);
321 #ifdef CONFIG_CGROUP_WRITEBACK
322 void wb_domain_exit(struct wb_domain *dom);
323 #endif
324
325 extern struct wb_domain global_wb_domain;
326
327 /* These are exported to sysctl. */
328 extern unsigned int dirty_writeback_interval;
329 extern unsigned int dirty_expire_interval;
330 extern unsigned int dirtytime_expire_interval;
331 extern int laptop_mode;
332
333 int dirtytime_interval_handler(const struct ctl_table *table, int write,
334 void *buffer, size_t *lenp, loff_t *ppos);
335
336 void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty);
337 unsigned long wb_calc_thresh(struct bdi_writeback *wb, unsigned long thresh);
338 unsigned long cgwb_calc_thresh(struct bdi_writeback *wb);
339
340 void wb_update_bandwidth(struct bdi_writeback *wb);
341
342 /* Invoke balance dirty pages in async mode. */
343 #define BDP_ASYNC 0x0001
344
345 void balance_dirty_pages_ratelimited(struct address_space *mapping);
346 int balance_dirty_pages_ratelimited_flags(struct address_space *mapping,
347 unsigned int flags);
348
349 bool wb_over_bg_thresh(struct bdi_writeback *wb);
350
351 struct folio *writeback_iter(struct address_space *mapping,
352 struct writeback_control *wbc, struct folio *folio, int *error);
353
354 typedef int (*writepage_t)(struct folio *folio, struct writeback_control *wbc,
355 void *data);
356
357 int write_cache_pages(struct address_space *mapping,
358 struct writeback_control *wbc, writepage_t writepage,
359 void *data);
360 int do_writepages(struct address_space *mapping, struct writeback_control *wbc);
361 void writeback_set_ratelimit(void);
362 void tag_pages_for_writeback(struct address_space *mapping,
363 pgoff_t start, pgoff_t end);
364
365 bool filemap_dirty_folio(struct address_space *mapping, struct folio *folio);
366 bool folio_redirty_for_writepage(struct writeback_control *, struct folio *);
367 bool redirty_page_for_writepage(struct writeback_control *, struct page *);
368
369 void sb_mark_inode_writeback(struct inode *inode);
370 void sb_clear_inode_writeback(struct inode *inode);
371
372 #endif /* WRITEBACK_H */
373