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 */
71 unsigned unpinned_fscache_wb:1; /* Cleared I_PINNING_FSCACHE_WB */
74 * When writeback IOs are bounced through async layers, only the
75 * initial synchronous phase should be accounted towards inode
76 * cgroup ownership arbitration to avoid confusion. Later stages
77 * can set the following flag to disable the accounting.
79 unsigned no_cgroup_owner:1;
81 unsigned punt_to_cgroup:1; /* cgrp punting, see __REQ_CGROUP_PUNT */
83 #ifdef CONFIG_CGROUP_WRITEBACK
84 struct bdi_writeback *wb; /* wb this writeback is issued under */
85 struct inode *inode; /* inode being written out */
87 /* foreign inode detection, see wbc_detach_inode() */
88 int wb_id; /* current wb id */
89 int wb_lcand_id; /* last foreign candidate wb id */
90 int wb_tcand_id; /* this foreign candidate wb id */
91 size_t wb_bytes; /* bytes written by current wb */
92 size_t wb_lcand_bytes; /* bytes written by last candidate */
93 size_t wb_tcand_bytes; /* bytes written by this candidate */
97 static inline int wbc_to_write_flags(struct writeback_control *wbc)
101 if (wbc->punt_to_cgroup)
102 flags = REQ_CGROUP_PUNT;
104 if (wbc->sync_mode == WB_SYNC_ALL)
106 else if (wbc->for_kupdate || wbc->for_background)
107 flags |= REQ_BACKGROUND;
112 #ifdef CONFIG_CGROUP_WRITEBACK
113 #define wbc_blkcg_css(wbc) \
114 ((wbc)->wb ? (wbc)->wb->blkcg_css : blkcg_root_css)
116 #define wbc_blkcg_css(wbc) (blkcg_root_css)
117 #endif /* CONFIG_CGROUP_WRITEBACK */
120 * A wb_domain represents a domain that wb's (bdi_writeback's) belong to
121 * and are measured against each other in. There always is one global
122 * domain, global_wb_domain, that every wb in the system is a member of.
123 * This allows measuring the relative bandwidth of each wb to distribute
124 * dirtyable memory accordingly.
130 * Scale the writeback cache size proportional to the relative
133 * We do this by keeping a floating proportion between BDIs, based
134 * on page writeback completions [end_page_writeback()]. Those
135 * devices that write out pages fastest will get the larger share,
136 * while the slower will get a smaller share.
138 * We use page writeout completions because we are interested in
139 * getting rid of dirty pages. Having them written out is the
142 * We introduce a concept of time, a period over which we measure
143 * these events, because demand can/will vary over time. The length
144 * of this period itself is measured in page writeback completions.
146 struct fprop_global completions;
147 struct timer_list period_timer; /* timer for aging of completions */
148 unsigned long period_time;
151 * The dirtyable memory and dirty threshold could be suddenly
152 * knocked down by a large amount (eg. on the startup of KVM in a
153 * swapless system). This may throw the system into deep dirty
154 * exceeded state and throttle heavy/light dirtiers alike. To
155 * retain good responsiveness, maintain global_dirty_limit for
156 * tracking slowly down to the knocked down dirty threshold.
158 * Both fields are protected by ->lock.
160 unsigned long dirty_limit_tstamp;
161 unsigned long dirty_limit;
165 * wb_domain_size_changed - memory available to a wb_domain has changed
166 * @dom: wb_domain of interest
168 * This function should be called when the amount of memory available to
169 * @dom has changed. It resets @dom's dirty limit parameters to prevent
170 * the past values which don't match the current configuration from skewing
171 * dirty throttling. Without this, when memory size of a wb_domain is
172 * greatly reduced, the dirty throttling logic may allow too many pages to
173 * be dirtied leading to consecutive unnecessary OOMs and may get stuck in
176 static inline void wb_domain_size_changed(struct wb_domain *dom)
178 spin_lock(&dom->lock);
179 dom->dirty_limit_tstamp = jiffies;
180 dom->dirty_limit = 0;
181 spin_unlock(&dom->lock);
187 struct bdi_writeback;
188 void writeback_inodes_sb(struct super_block *, enum wb_reason reason);
189 void writeback_inodes_sb_nr(struct super_block *, unsigned long nr,
190 enum wb_reason reason);
191 void try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason);
192 void sync_inodes_sb(struct super_block *);
193 void wakeup_flusher_threads(enum wb_reason reason);
194 void wakeup_flusher_threads_bdi(struct backing_dev_info *bdi,
195 enum wb_reason reason);
196 void inode_wait_for_writeback(struct inode *inode);
197 void inode_io_list_del(struct inode *inode);
199 /* writeback.h requires fs.h; it, too, is not included from here. */
200 static inline void wait_on_inode(struct inode *inode)
203 wait_on_bit(&inode->i_state, __I_NEW, TASK_UNINTERRUPTIBLE);
206 #ifdef CONFIG_CGROUP_WRITEBACK
208 #include <linux/cgroup.h>
209 #include <linux/bio.h>
211 void __inode_attach_wb(struct inode *inode, struct page *page);
212 void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
214 __releases(&inode->i_lock);
215 void wbc_detach_inode(struct writeback_control *wbc);
216 void wbc_account_cgroup_owner(struct writeback_control *wbc, struct page *page,
218 int cgroup_writeback_by_id(u64 bdi_id, int memcg_id,
219 enum wb_reason reason, struct wb_completion *done);
220 void cgroup_writeback_umount(void);
221 bool cleanup_offline_cgwb(struct bdi_writeback *wb);
224 * inode_attach_wb - associate an inode with its wb
225 * @inode: inode of interest
226 * @page: page being dirtied (may be NULL)
228 * If @inode doesn't have its wb, associate it with the wb matching the
229 * memcg of @page or, if @page is NULL, %current. May be called w/ or w/o
232 static inline void inode_attach_wb(struct inode *inode, struct page *page)
235 __inode_attach_wb(inode, page);
239 * inode_detach_wb - disassociate an inode from its wb
240 * @inode: inode of interest
242 * @inode is being freed. Detach from its wb.
244 static inline void inode_detach_wb(struct inode *inode)
247 WARN_ON_ONCE(!(inode->i_state & I_CLEAR));
254 * wbc_attach_fdatawrite_inode - associate wbc and inode for fdatawrite
255 * @wbc: writeback_control of interest
256 * @inode: target inode
258 * This function is to be used by __filemap_fdatawrite_range(), which is an
259 * alternative entry point into writeback code, and first ensures @inode is
260 * associated with a bdi_writeback and attaches it to @wbc.
262 static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
265 spin_lock(&inode->i_lock);
266 inode_attach_wb(inode, NULL);
267 wbc_attach_and_unlock_inode(wbc, inode);
271 * wbc_init_bio - writeback specific initializtion of bio
272 * @wbc: writeback_control for the writeback in progress
273 * @bio: bio to be initialized
275 * @bio is a part of the writeback in progress controlled by @wbc. Perform
276 * writeback specific initialization. This is used to apply the cgroup
277 * writeback context. Must be called after the bio has been associated with
280 static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
283 * pageout() path doesn't attach @wbc to the inode being written
284 * out. This is intentional as we don't want the function to block
285 * behind a slow cgroup. Ultimately, we want pageout() to kick off
286 * regular writeback instead of writing things out itself.
289 bio_associate_blkg_from_css(bio, wbc->wb->blkcg_css);
292 #else /* CONFIG_CGROUP_WRITEBACK */
294 static inline void inode_attach_wb(struct inode *inode, struct page *page)
298 static inline void inode_detach_wb(struct inode *inode)
302 static inline void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
304 __releases(&inode->i_lock)
306 spin_unlock(&inode->i_lock);
309 static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
314 static inline void wbc_detach_inode(struct writeback_control *wbc)
318 static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
322 static inline void wbc_account_cgroup_owner(struct writeback_control *wbc,
323 struct page *page, size_t bytes)
327 static inline void cgroup_writeback_umount(void)
331 #endif /* CONFIG_CGROUP_WRITEBACK */
334 * mm/page-writeback.c
336 void laptop_io_completion(struct backing_dev_info *info);
337 void laptop_sync_completion(void);
338 void laptop_mode_timer_fn(struct timer_list *t);
339 bool node_dirty_ok(struct pglist_data *pgdat);
340 int wb_domain_init(struct wb_domain *dom, gfp_t gfp);
341 #ifdef CONFIG_CGROUP_WRITEBACK
342 void wb_domain_exit(struct wb_domain *dom);
345 extern struct wb_domain global_wb_domain;
347 /* These are exported to sysctl. */
348 extern int dirty_background_ratio;
349 extern unsigned long dirty_background_bytes;
350 extern int vm_dirty_ratio;
351 extern unsigned long vm_dirty_bytes;
352 extern unsigned int dirty_writeback_interval;
353 extern unsigned int dirty_expire_interval;
354 extern unsigned int dirtytime_expire_interval;
355 extern int vm_highmem_is_dirtyable;
356 extern int laptop_mode;
358 int dirty_background_ratio_handler(struct ctl_table *table, int write,
359 void *buffer, size_t *lenp, loff_t *ppos);
360 int dirty_background_bytes_handler(struct ctl_table *table, int write,
361 void *buffer, size_t *lenp, loff_t *ppos);
362 int dirty_ratio_handler(struct ctl_table *table, int write,
363 void *buffer, size_t *lenp, loff_t *ppos);
364 int dirty_bytes_handler(struct ctl_table *table, int write,
365 void *buffer, size_t *lenp, loff_t *ppos);
366 int dirtytime_interval_handler(struct ctl_table *table, int write,
367 void *buffer, size_t *lenp, loff_t *ppos);
368 int dirty_writeback_centisecs_handler(struct ctl_table *table, int write,
369 void *buffer, size_t *lenp, loff_t *ppos);
371 void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty);
372 unsigned long wb_calc_thresh(struct bdi_writeback *wb, unsigned long thresh);
374 void wb_update_bandwidth(struct bdi_writeback *wb);
375 void balance_dirty_pages_ratelimited(struct address_space *mapping);
376 bool wb_over_bg_thresh(struct bdi_writeback *wb);
378 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
381 int generic_writepages(struct address_space *mapping,
382 struct writeback_control *wbc);
383 void tag_pages_for_writeback(struct address_space *mapping,
384 pgoff_t start, pgoff_t end);
385 int write_cache_pages(struct address_space *mapping,
386 struct writeback_control *wbc, writepage_t writepage,
388 int do_writepages(struct address_space *mapping, struct writeback_control *wbc);
389 void writeback_set_ratelimit(void);
390 void tag_pages_for_writeback(struct address_space *mapping,
391 pgoff_t start, pgoff_t end);
393 bool filemap_dirty_folio(struct address_space *mapping, struct folio *folio);
394 void folio_account_redirty(struct folio *folio);
395 static inline void account_page_redirty(struct page *page)
397 folio_account_redirty(page_folio(page));
399 bool folio_redirty_for_writepage(struct writeback_control *, struct folio *);
400 bool redirty_page_for_writepage(struct writeback_control *, struct page *);
402 void sb_mark_inode_writeback(struct inode *inode);
403 void sb_clear_inode_writeback(struct inode *inode);
405 #endif /* WRITEBACK_H */