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>
14 #include <linux/blk-cgroup.h>
18 DECLARE_PER_CPU(int, dirty_throttle_leaks);
21 * The 1/4 region under the global dirty thresh is for smooth dirty throttling:
23 * (thresh - thresh/DIRTY_FULL_SCOPE, thresh)
25 * Further beyond, all dirtier tasks will enter a loop waiting (possibly long
26 * time) for the dirty pages to drop, unless written enough pages.
28 * The global dirty threshold is normally equal to the global dirty limit,
29 * except when the system suddenly allocates a lot of anonymous memory and
30 * knocks down the global dirty threshold quickly, in which case the global
31 * dirty limit will follow down slowly to prevent livelocking all dirtier tasks.
34 #define DIRTY_FULL_SCOPE (DIRTY_SCOPE / 2)
36 struct backing_dev_info;
41 enum writeback_sync_modes {
42 WB_SYNC_NONE, /* Don't wait on anything */
43 WB_SYNC_ALL, /* Wait on every mapping */
47 * A control structure which tells the writeback code what to do. These are
48 * always on the stack, and hence need no locking. They are always initialised
49 * in a manner such that unspecified fields are set to zero.
51 struct writeback_control {
52 long nr_to_write; /* Write this many pages, and decrement
53 this for each page written */
54 long pages_skipped; /* Pages which were not written */
57 * For a_ops->writepages(): if start or end are non-zero then this is
58 * a hint that the filesystem need only write out the pages inside that
59 * byterange. The byte at `end' is included in the writeout request.
64 enum writeback_sync_modes sync_mode;
66 unsigned for_kupdate:1; /* A kupdate writeback */
67 unsigned for_background:1; /* A background writeback */
68 unsigned tagged_writepages:1; /* tag-and-write to avoid livelock */
69 unsigned for_reclaim:1; /* Invoked from the page allocator */
70 unsigned range_cyclic:1; /* range_start is cyclic */
71 unsigned for_sync:1; /* sync(2) WB_SYNC_ALL writeback */
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 #ifdef CONFIG_CGROUP_WRITEBACK
82 struct bdi_writeback *wb; /* wb this writeback is issued under */
83 struct inode *inode; /* inode being written out */
85 /* foreign inode detection, see wbc_detach_inode() */
86 int wb_id; /* current wb id */
87 int wb_lcand_id; /* last foreign candidate wb id */
88 int wb_tcand_id; /* this foreign candidate wb id */
89 size_t wb_bytes; /* bytes written by current wb */
90 size_t wb_lcand_bytes; /* bytes written by last candidate */
91 size_t wb_tcand_bytes; /* bytes written by this candidate */
95 static inline int wbc_to_write_flags(struct writeback_control *wbc)
97 if (wbc->sync_mode == WB_SYNC_ALL)
99 else if (wbc->for_kupdate || wbc->for_background)
100 return REQ_BACKGROUND;
105 static inline struct cgroup_subsys_state *
106 wbc_blkcg_css(struct writeback_control *wbc)
108 #ifdef CONFIG_CGROUP_WRITEBACK
110 return wbc->wb->blkcg_css;
112 return blkcg_root_css;
116 * A wb_domain represents a domain that wb's (bdi_writeback's) belong to
117 * and are measured against each other in. There always is one global
118 * domain, global_wb_domain, that every wb in the system is a member of.
119 * This allows measuring the relative bandwidth of each wb to distribute
120 * dirtyable memory accordingly.
126 * Scale the writeback cache size proportional to the relative
129 * We do this by keeping a floating proportion between BDIs, based
130 * on page writeback completions [end_page_writeback()]. Those
131 * devices that write out pages fastest will get the larger share,
132 * while the slower will get a smaller share.
134 * We use page writeout completions because we are interested in
135 * getting rid of dirty pages. Having them written out is the
138 * We introduce a concept of time, a period over which we measure
139 * these events, because demand can/will vary over time. The length
140 * of this period itself is measured in page writeback completions.
142 struct fprop_global completions;
143 struct timer_list period_timer; /* timer for aging of completions */
144 unsigned long period_time;
147 * The dirtyable memory and dirty threshold could be suddenly
148 * knocked down by a large amount (eg. on the startup of KVM in a
149 * swapless system). This may throw the system into deep dirty
150 * exceeded state and throttle heavy/light dirtiers alike. To
151 * retain good responsiveness, maintain global_dirty_limit for
152 * tracking slowly down to the knocked down dirty threshold.
154 * Both fields are protected by ->lock.
156 unsigned long dirty_limit_tstamp;
157 unsigned long dirty_limit;
161 * wb_domain_size_changed - memory available to a wb_domain has changed
162 * @dom: wb_domain of interest
164 * This function should be called when the amount of memory available to
165 * @dom has changed. It resets @dom's dirty limit parameters to prevent
166 * the past values which don't match the current configuration from skewing
167 * dirty throttling. Without this, when memory size of a wb_domain is
168 * greatly reduced, the dirty throttling logic may allow too many pages to
169 * be dirtied leading to consecutive unnecessary OOMs and may get stuck in
172 static inline void wb_domain_size_changed(struct wb_domain *dom)
174 spin_lock(&dom->lock);
175 dom->dirty_limit_tstamp = jiffies;
176 dom->dirty_limit = 0;
177 spin_unlock(&dom->lock);
183 struct bdi_writeback;
184 void writeback_inodes_sb(struct super_block *, enum wb_reason reason);
185 void writeback_inodes_sb_nr(struct super_block *, unsigned long nr,
186 enum wb_reason reason);
187 void try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason);
188 void sync_inodes_sb(struct super_block *);
189 void wakeup_flusher_threads(enum wb_reason reason);
190 void wakeup_flusher_threads_bdi(struct backing_dev_info *bdi,
191 enum wb_reason reason);
192 void inode_wait_for_writeback(struct inode *inode);
194 /* writeback.h requires fs.h; it, too, is not included from here. */
195 static inline void wait_on_inode(struct inode *inode)
198 wait_on_bit(&inode->i_state, __I_NEW, TASK_UNINTERRUPTIBLE);
201 #ifdef CONFIG_CGROUP_WRITEBACK
203 #include <linux/cgroup.h>
204 #include <linux/bio.h>
206 void __inode_attach_wb(struct inode *inode, struct page *page);
207 void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
209 __releases(&inode->i_lock);
210 void wbc_detach_inode(struct writeback_control *wbc);
211 void wbc_account_cgroup_owner(struct writeback_control *wbc, struct page *page,
213 void cgroup_writeback_umount(void);
216 * inode_attach_wb - associate an inode with its wb
217 * @inode: inode of interest
218 * @page: page being dirtied (may be NULL)
220 * If @inode doesn't have its wb, associate it with the wb matching the
221 * memcg of @page or, if @page is NULL, %current. May be called w/ or w/o
224 static inline void inode_attach_wb(struct inode *inode, struct page *page)
227 __inode_attach_wb(inode, page);
231 * inode_detach_wb - disassociate an inode from its wb
232 * @inode: inode of interest
234 * @inode is being freed. Detach from its wb.
236 static inline void inode_detach_wb(struct inode *inode)
239 WARN_ON_ONCE(!(inode->i_state & I_CLEAR));
246 * wbc_attach_fdatawrite_inode - associate wbc and inode for fdatawrite
247 * @wbc: writeback_control of interest
248 * @inode: target inode
250 * This function is to be used by __filemap_fdatawrite_range(), which is an
251 * alternative entry point into writeback code, and first ensures @inode is
252 * associated with a bdi_writeback and attaches it to @wbc.
254 static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
257 spin_lock(&inode->i_lock);
258 inode_attach_wb(inode, NULL);
259 wbc_attach_and_unlock_inode(wbc, inode);
263 * wbc_init_bio - writeback specific initializtion of bio
264 * @wbc: writeback_control for the writeback in progress
265 * @bio: bio to be initialized
267 * @bio is a part of the writeback in progress controlled by @wbc. Perform
268 * writeback specific initialization. This is used to apply the cgroup
269 * writeback context. Must be called after the bio has been associated with
272 static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
275 * pageout() path doesn't attach @wbc to the inode being written
276 * out. This is intentional as we don't want the function to block
277 * behind a slow cgroup. Ultimately, we want pageout() to kick off
278 * regular writeback instead of writing things out itself.
281 bio_associate_blkg_from_css(bio, wbc->wb->blkcg_css);
284 #else /* CONFIG_CGROUP_WRITEBACK */
286 static inline void inode_attach_wb(struct inode *inode, struct page *page)
290 static inline void inode_detach_wb(struct inode *inode)
294 static inline void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
296 __releases(&inode->i_lock)
298 spin_unlock(&inode->i_lock);
301 static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
306 static inline void wbc_detach_inode(struct writeback_control *wbc)
310 static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
314 static inline void wbc_account_cgroup_owner(struct writeback_control *wbc,
315 struct page *page, size_t bytes)
319 static inline void cgroup_writeback_umount(void)
323 #endif /* CONFIG_CGROUP_WRITEBACK */
326 * mm/page-writeback.c
329 void laptop_io_completion(struct backing_dev_info *info);
330 void laptop_sync_completion(void);
331 void laptop_mode_sync(struct work_struct *work);
332 void laptop_mode_timer_fn(struct timer_list *t);
334 static inline void laptop_sync_completion(void) { }
336 bool node_dirty_ok(struct pglist_data *pgdat);
337 int wb_domain_init(struct wb_domain *dom, gfp_t gfp);
338 #ifdef CONFIG_CGROUP_WRITEBACK
339 void wb_domain_exit(struct wb_domain *dom);
342 extern struct wb_domain global_wb_domain;
344 /* These are exported to sysctl. */
345 extern int dirty_background_ratio;
346 extern unsigned long dirty_background_bytes;
347 extern int vm_dirty_ratio;
348 extern unsigned long vm_dirty_bytes;
349 extern unsigned int dirty_writeback_interval;
350 extern unsigned int dirty_expire_interval;
351 extern unsigned int dirtytime_expire_interval;
352 extern int vm_highmem_is_dirtyable;
353 extern int block_dump;
354 extern int laptop_mode;
356 extern int dirty_background_ratio_handler(struct ctl_table *table, int write,
357 void __user *buffer, size_t *lenp,
359 extern int dirty_background_bytes_handler(struct ctl_table *table, int write,
360 void __user *buffer, size_t *lenp,
362 extern int dirty_ratio_handler(struct ctl_table *table, int write,
363 void __user *buffer, size_t *lenp,
365 extern int dirty_bytes_handler(struct ctl_table *table, int write,
366 void __user *buffer, size_t *lenp,
368 int dirtytime_interval_handler(struct ctl_table *table, int write,
369 void __user *buffer, size_t *lenp, loff_t *ppos);
372 int dirty_writeback_centisecs_handler(struct ctl_table *, int,
373 void __user *, size_t *, loff_t *);
375 void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty);
376 unsigned long wb_calc_thresh(struct bdi_writeback *wb, unsigned long thresh);
378 void wb_update_bandwidth(struct bdi_writeback *wb, unsigned long start_time);
379 void balance_dirty_pages_ratelimited(struct address_space *mapping);
380 bool wb_over_bg_thresh(struct bdi_writeback *wb);
382 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
385 int generic_writepages(struct address_space *mapping,
386 struct writeback_control *wbc);
387 void tag_pages_for_writeback(struct address_space *mapping,
388 pgoff_t start, pgoff_t end);
389 int write_cache_pages(struct address_space *mapping,
390 struct writeback_control *wbc, writepage_t writepage,
392 int do_writepages(struct address_space *mapping, struct writeback_control *wbc);
393 void writeback_set_ratelimit(void);
394 void tag_pages_for_writeback(struct address_space *mapping,
395 pgoff_t start, pgoff_t end);
397 void account_page_redirty(struct page *page);
399 void sb_mark_inode_writeback(struct inode *inode);
400 void sb_clear_inode_writeback(struct inode *inode);
402 #endif /* WRITEBACK_H */