1 /* SPDX-License-Identifier: GPL-2.0 */
3 * include/linux/writeback.h
8 #include <linux/sched.h>
9 #include <linux/workqueue.h>
11 #include <linux/flex_proportions.h>
12 #include <linux/backing-dev-defs.h>
13 #include <linux/blk_types.h>
17 DECLARE_PER_CPU(int, dirty_throttle_leaks);
20 * The 1/4 region under the global dirty thresh is for smooth dirty throttling:
22 * (thresh - thresh/DIRTY_FULL_SCOPE, thresh)
24 * Further beyond, all dirtier tasks will enter a loop waiting (possibly long
25 * time) for the dirty pages to drop, unless written enough pages.
27 * The global dirty threshold is normally equal to the global dirty limit,
28 * except when the system suddenly allocates a lot of anonymous memory and
29 * knocks down the global dirty threshold quickly, in which case the global
30 * dirty limit will follow down slowly to prevent livelocking all dirtier tasks.
33 #define DIRTY_FULL_SCOPE (DIRTY_SCOPE / 2)
35 struct backing_dev_info;
40 enum writeback_sync_modes {
41 WB_SYNC_NONE, /* Don't wait on anything */
42 WB_SYNC_ALL, /* Wait on every mapping */
46 * A control structure which tells the writeback code what to do. These are
47 * always on the stack, and hence need no locking. They are always initialised
48 * in a manner such that unspecified fields are set to zero.
50 struct writeback_control {
51 long nr_to_write; /* Write this many pages, and decrement
52 this for each page written */
53 long pages_skipped; /* Pages which were not written */
56 * For a_ops->writepages(): if start or end are non-zero then this is
57 * a hint that the filesystem need only write out the pages inside that
58 * byterange. The byte at `end' is included in the writeout request.
63 enum writeback_sync_modes sync_mode;
65 unsigned for_kupdate:1; /* A kupdate writeback */
66 unsigned for_background:1; /* A background writeback */
67 unsigned tagged_writepages:1; /* tag-and-write to avoid livelock */
68 unsigned for_reclaim:1; /* Invoked from the page allocator */
69 unsigned range_cyclic:1; /* range_start is cyclic */
70 unsigned for_sync:1; /* sync(2) WB_SYNC_ALL writeback */
73 * When writeback IOs are bounced through async layers, only the
74 * initial synchronous phase should be accounted towards inode
75 * cgroup ownership arbitration to avoid confusion. Later stages
76 * can set the following flag to disable the accounting.
78 unsigned no_cgroup_owner:1;
80 unsigned punt_to_cgroup:1; /* cgrp punting, see __REQ_CGROUP_PUNT */
82 #ifdef CONFIG_CGROUP_WRITEBACK
83 struct bdi_writeback *wb; /* wb this writeback is issued under */
84 struct inode *inode; /* inode being written out */
86 /* foreign inode detection, see wbc_detach_inode() */
87 int wb_id; /* current wb id */
88 int wb_lcand_id; /* last foreign candidate wb id */
89 int wb_tcand_id; /* this foreign candidate wb id */
90 size_t wb_bytes; /* bytes written by current wb */
91 size_t wb_lcand_bytes; /* bytes written by last candidate */
92 size_t wb_tcand_bytes; /* bytes written by this candidate */
96 static inline int wbc_to_write_flags(struct writeback_control *wbc)
100 if (wbc->punt_to_cgroup)
101 flags = REQ_CGROUP_PUNT;
103 if (wbc->sync_mode == WB_SYNC_ALL)
105 else if (wbc->for_kupdate || wbc->for_background)
106 flags |= REQ_BACKGROUND;
111 #ifdef CONFIG_CGROUP_WRITEBACK
112 #define wbc_blkcg_css(wbc) \
113 ((wbc)->wb ? (wbc)->wb->blkcg_css : blkcg_root_css)
115 #define wbc_blkcg_css(wbc) (blkcg_root_css)
116 #endif /* CONFIG_CGROUP_WRITEBACK */
119 * A wb_domain represents a domain that wb's (bdi_writeback's) belong to
120 * and are measured against each other in. There always is one global
121 * domain, global_wb_domain, that every wb in the system is a member of.
122 * This allows measuring the relative bandwidth of each wb to distribute
123 * dirtyable memory accordingly.
129 * Scale the writeback cache size proportional to the relative
132 * We do this by keeping a floating proportion between BDIs, based
133 * on page writeback completions [end_page_writeback()]. Those
134 * devices that write out pages fastest will get the larger share,
135 * while the slower will get a smaller share.
137 * We use page writeout completions because we are interested in
138 * getting rid of dirty pages. Having them written out is the
141 * We introduce a concept of time, a period over which we measure
142 * these events, because demand can/will vary over time. The length
143 * of this period itself is measured in page writeback completions.
145 struct fprop_global completions;
146 struct timer_list period_timer; /* timer for aging of completions */
147 unsigned long period_time;
150 * The dirtyable memory and dirty threshold could be suddenly
151 * knocked down by a large amount (eg. on the startup of KVM in a
152 * swapless system). This may throw the system into deep dirty
153 * exceeded state and throttle heavy/light dirtiers alike. To
154 * retain good responsiveness, maintain global_dirty_limit for
155 * tracking slowly down to the knocked down dirty threshold.
157 * Both fields are protected by ->lock.
159 unsigned long dirty_limit_tstamp;
160 unsigned long dirty_limit;
164 * wb_domain_size_changed - memory available to a wb_domain has changed
165 * @dom: wb_domain of interest
167 * This function should be called when the amount of memory available to
168 * @dom has changed. It resets @dom's dirty limit parameters to prevent
169 * the past values which don't match the current configuration from skewing
170 * dirty throttling. Without this, when memory size of a wb_domain is
171 * greatly reduced, the dirty throttling logic may allow too many pages to
172 * be dirtied leading to consecutive unnecessary OOMs and may get stuck in
175 static inline void wb_domain_size_changed(struct wb_domain *dom)
177 spin_lock(&dom->lock);
178 dom->dirty_limit_tstamp = jiffies;
179 dom->dirty_limit = 0;
180 spin_unlock(&dom->lock);
186 struct bdi_writeback;
187 void writeback_inodes_sb(struct super_block *, enum wb_reason reason);
188 void writeback_inodes_sb_nr(struct super_block *, unsigned long nr,
189 enum wb_reason reason);
190 void try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason);
191 void sync_inodes_sb(struct super_block *);
192 void wakeup_flusher_threads(enum wb_reason reason);
193 void wakeup_flusher_threads_bdi(struct backing_dev_info *bdi,
194 enum wb_reason reason);
195 void inode_wait_for_writeback(struct inode *inode);
196 void inode_io_list_del(struct inode *inode);
198 /* writeback.h requires fs.h; it, too, is not included from here. */
199 static inline void wait_on_inode(struct inode *inode)
202 wait_on_bit(&inode->i_state, __I_NEW, TASK_UNINTERRUPTIBLE);
205 #ifdef CONFIG_CGROUP_WRITEBACK
207 #include <linux/cgroup.h>
208 #include <linux/bio.h>
210 void __inode_attach_wb(struct inode *inode, struct page *page);
211 void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
213 __releases(&inode->i_lock);
214 void wbc_detach_inode(struct writeback_control *wbc);
215 void wbc_account_cgroup_owner(struct writeback_control *wbc, struct page *page,
217 int cgroup_writeback_by_id(u64 bdi_id, int memcg_id,
218 enum wb_reason reason, struct wb_completion *done);
219 void cgroup_writeback_umount(void);
220 bool cleanup_offline_cgwb(struct bdi_writeback *wb);
223 * inode_attach_wb - associate an inode with its wb
224 * @inode: inode of interest
225 * @page: page being dirtied (may be NULL)
227 * If @inode doesn't have its wb, associate it with the wb matching the
228 * memcg of @page or, if @page is NULL, %current. May be called w/ or w/o
231 static inline void inode_attach_wb(struct inode *inode, struct page *page)
234 __inode_attach_wb(inode, page);
238 * inode_detach_wb - disassociate an inode from its wb
239 * @inode: inode of interest
241 * @inode is being freed. Detach from its wb.
243 static inline void inode_detach_wb(struct inode *inode)
246 WARN_ON_ONCE(!(inode->i_state & I_CLEAR));
253 * wbc_attach_fdatawrite_inode - associate wbc and inode for fdatawrite
254 * @wbc: writeback_control of interest
255 * @inode: target inode
257 * This function is to be used by __filemap_fdatawrite_range(), which is an
258 * alternative entry point into writeback code, and first ensures @inode is
259 * associated with a bdi_writeback and attaches it to @wbc.
261 static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
264 spin_lock(&inode->i_lock);
265 inode_attach_wb(inode, NULL);
266 wbc_attach_and_unlock_inode(wbc, inode);
270 * wbc_init_bio - writeback specific initializtion of bio
271 * @wbc: writeback_control for the writeback in progress
272 * @bio: bio to be initialized
274 * @bio is a part of the writeback in progress controlled by @wbc. Perform
275 * writeback specific initialization. This is used to apply the cgroup
276 * writeback context. Must be called after the bio has been associated with
279 static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
282 * pageout() path doesn't attach @wbc to the inode being written
283 * out. This is intentional as we don't want the function to block
284 * behind a slow cgroup. Ultimately, we want pageout() to kick off
285 * regular writeback instead of writing things out itself.
288 bio_associate_blkg_from_css(bio, wbc->wb->blkcg_css);
291 #else /* CONFIG_CGROUP_WRITEBACK */
293 static inline void inode_attach_wb(struct inode *inode, struct page *page)
297 static inline void inode_detach_wb(struct inode *inode)
301 static inline void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
303 __releases(&inode->i_lock)
305 spin_unlock(&inode->i_lock);
308 static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
313 static inline void wbc_detach_inode(struct writeback_control *wbc)
317 static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
321 static inline void wbc_account_cgroup_owner(struct writeback_control *wbc,
322 struct page *page, size_t bytes)
326 static inline void cgroup_writeback_umount(void)
330 #endif /* CONFIG_CGROUP_WRITEBACK */
333 * mm/page-writeback.c
335 void laptop_io_completion(struct backing_dev_info *info);
336 void laptop_sync_completion(void);
337 void laptop_mode_timer_fn(struct timer_list *t);
338 bool node_dirty_ok(struct pglist_data *pgdat);
339 int wb_domain_init(struct wb_domain *dom, gfp_t gfp);
340 #ifdef CONFIG_CGROUP_WRITEBACK
341 void wb_domain_exit(struct wb_domain *dom);
344 extern struct wb_domain global_wb_domain;
346 /* These are exported to sysctl. */
347 extern int dirty_background_ratio;
348 extern unsigned long dirty_background_bytes;
349 extern int vm_dirty_ratio;
350 extern unsigned long vm_dirty_bytes;
351 extern unsigned int dirty_writeback_interval;
352 extern unsigned int dirty_expire_interval;
353 extern unsigned int dirtytime_expire_interval;
354 extern int vm_highmem_is_dirtyable;
355 extern int laptop_mode;
357 int dirty_background_ratio_handler(struct ctl_table *table, int write,
358 void *buffer, size_t *lenp, loff_t *ppos);
359 int dirty_background_bytes_handler(struct ctl_table *table, int write,
360 void *buffer, size_t *lenp, loff_t *ppos);
361 int dirty_ratio_handler(struct ctl_table *table, int write,
362 void *buffer, size_t *lenp, loff_t *ppos);
363 int dirty_bytes_handler(struct ctl_table *table, int write,
364 void *buffer, size_t *lenp, loff_t *ppos);
365 int dirtytime_interval_handler(struct ctl_table *table, int write,
366 void *buffer, size_t *lenp, loff_t *ppos);
367 int dirty_writeback_centisecs_handler(struct ctl_table *table, int write,
368 void *buffer, size_t *lenp, loff_t *ppos);
370 void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty);
371 unsigned long wb_calc_thresh(struct bdi_writeback *wb, unsigned long thresh);
373 void wb_update_bandwidth(struct bdi_writeback *wb);
374 void balance_dirty_pages_ratelimited(struct address_space *mapping);
375 bool wb_over_bg_thresh(struct bdi_writeback *wb);
377 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
380 int generic_writepages(struct address_space *mapping,
381 struct writeback_control *wbc);
382 void tag_pages_for_writeback(struct address_space *mapping,
383 pgoff_t start, pgoff_t end);
384 int write_cache_pages(struct address_space *mapping,
385 struct writeback_control *wbc, writepage_t writepage,
387 int do_writepages(struct address_space *mapping, struct writeback_control *wbc);
388 void writeback_set_ratelimit(void);
389 void tag_pages_for_writeback(struct address_space *mapping,
390 pgoff_t start, pgoff_t end);
392 bool filemap_dirty_folio(struct address_space *mapping, struct folio *folio);
393 void folio_account_redirty(struct folio *folio);
394 static inline void account_page_redirty(struct page *page)
396 folio_account_redirty(page_folio(page));
398 bool folio_redirty_for_writepage(struct writeback_control *, struct folio *);
399 bool redirty_page_for_writepage(struct writeback_control *, struct page *);
401 void sb_mark_inode_writeback(struct inode *inode);
402 void sb_clear_inode_writeback(struct inode *inode);
404 #endif /* WRITEBACK_H */