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 * why some writeback work was initiated
53 WB_REASON_LAPTOP_TIMER,
54 WB_REASON_FREE_MORE_MEM,
55 WB_REASON_FS_FREE_SPACE,
57 * There is no bdi forker thread any more and works are done
58 * by emergency worker, however, this is TPs userland visible
59 * and we'll be exposing exactly the same information,
60 * so it has a mismatch name.
62 WB_REASON_FORKER_THREAD,
68 * A control structure which tells the writeback code what to do. These are
69 * always on the stack, and hence need no locking. They are always initialised
70 * in a manner such that unspecified fields are set to zero.
72 struct writeback_control {
73 long nr_to_write; /* Write this many pages, and decrement
74 this for each page written */
75 long pages_skipped; /* Pages which were not written */
78 * For a_ops->writepages(): if start or end are non-zero then this is
79 * a hint that the filesystem need only write out the pages inside that
80 * byterange. The byte at `end' is included in the writeout request.
85 enum writeback_sync_modes sync_mode;
87 unsigned for_kupdate:1; /* A kupdate writeback */
88 unsigned for_background:1; /* A background writeback */
89 unsigned tagged_writepages:1; /* tag-and-write to avoid livelock */
90 unsigned for_reclaim:1; /* Invoked from the page allocator */
91 unsigned range_cyclic:1; /* range_start is cyclic */
92 unsigned for_sync:1; /* sync(2) WB_SYNC_ALL writeback */
93 #ifdef CONFIG_CGROUP_WRITEBACK
94 struct bdi_writeback *wb; /* wb this writeback is issued under */
95 struct inode *inode; /* inode being written out */
97 /* foreign inode detection, see wbc_detach_inode() */
98 int wb_id; /* current wb id */
99 int wb_lcand_id; /* last foreign candidate wb id */
100 int wb_tcand_id; /* this foreign candidate wb id */
101 size_t wb_bytes; /* bytes written by current wb */
102 size_t wb_lcand_bytes; /* bytes written by last candidate */
103 size_t wb_tcand_bytes; /* bytes written by this candidate */
107 static inline int wbc_to_write_flags(struct writeback_control *wbc)
109 if (wbc->sync_mode == WB_SYNC_ALL)
111 else if (wbc->for_kupdate || wbc->for_background)
112 return REQ_BACKGROUND;
118 * A wb_domain represents a domain that wb's (bdi_writeback's) belong to
119 * and are measured against each other in. There always is one global
120 * domain, global_wb_domain, that every wb in the system is a member of.
121 * This allows measuring the relative bandwidth of each wb to distribute
122 * dirtyable memory accordingly.
128 * Scale the writeback cache size proportional to the relative
131 * We do this by keeping a floating proportion between BDIs, based
132 * on page writeback completions [end_page_writeback()]. Those
133 * devices that write out pages fastest will get the larger share,
134 * while the slower will get a smaller share.
136 * We use page writeout completions because we are interested in
137 * getting rid of dirty pages. Having them written out is the
140 * We introduce a concept of time, a period over which we measure
141 * these events, because demand can/will vary over time. The length
142 * of this period itself is measured in page writeback completions.
144 struct fprop_global completions;
145 struct timer_list period_timer; /* timer for aging of completions */
146 unsigned long period_time;
149 * The dirtyable memory and dirty threshold could be suddenly
150 * knocked down by a large amount (eg. on the startup of KVM in a
151 * swapless system). This may throw the system into deep dirty
152 * exceeded state and throttle heavy/light dirtiers alike. To
153 * retain good responsiveness, maintain global_dirty_limit for
154 * tracking slowly down to the knocked down dirty threshold.
156 * Both fields are protected by ->lock.
158 unsigned long dirty_limit_tstamp;
159 unsigned long dirty_limit;
163 * wb_domain_size_changed - memory available to a wb_domain has changed
164 * @dom: wb_domain of interest
166 * This function should be called when the amount of memory available to
167 * @dom has changed. It resets @dom's dirty limit parameters to prevent
168 * the past values which don't match the current configuration from skewing
169 * dirty throttling. Without this, when memory size of a wb_domain is
170 * greatly reduced, the dirty throttling logic may allow too many pages to
171 * be dirtied leading to consecutive unnecessary OOMs and may get stuck in
174 static inline void wb_domain_size_changed(struct wb_domain *dom)
176 spin_lock(&dom->lock);
177 dom->dirty_limit_tstamp = jiffies;
178 dom->dirty_limit = 0;
179 spin_unlock(&dom->lock);
185 struct bdi_writeback;
186 void writeback_inodes_sb(struct super_block *, enum wb_reason reason);
187 void writeback_inodes_sb_nr(struct super_block *, unsigned long nr,
188 enum wb_reason reason);
189 bool try_to_writeback_inodes_sb(struct super_block *, enum wb_reason reason);
190 bool try_to_writeback_inodes_sb_nr(struct super_block *, unsigned long nr,
191 enum wb_reason reason);
192 void sync_inodes_sb(struct super_block *);
193 void wakeup_flusher_threads(long nr_pages, enum wb_reason reason);
194 void inode_wait_for_writeback(struct inode *inode);
196 /* writeback.h requires fs.h; it, too, is not included from here. */
197 static inline void wait_on_inode(struct inode *inode)
200 wait_on_bit(&inode->i_state, __I_NEW, TASK_UNINTERRUPTIBLE);
203 #ifdef CONFIG_CGROUP_WRITEBACK
205 #include <linux/cgroup.h>
206 #include <linux/bio.h>
208 void __inode_attach_wb(struct inode *inode, struct page *page);
209 void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
211 __releases(&inode->i_lock);
212 void wbc_detach_inode(struct writeback_control *wbc);
213 void wbc_account_io(struct writeback_control *wbc, struct page *page,
215 void cgroup_writeback_umount(void);
218 * inode_attach_wb - associate an inode with its wb
219 * @inode: inode of interest
220 * @page: page being dirtied (may be NULL)
222 * If @inode doesn't have its wb, associate it with the wb matching the
223 * memcg of @page or, if @page is NULL, %current. May be called w/ or w/o
226 static inline void inode_attach_wb(struct inode *inode, struct page *page)
229 __inode_attach_wb(inode, page);
233 * inode_detach_wb - disassociate an inode from its wb
234 * @inode: inode of interest
236 * @inode is being freed. Detach from its wb.
238 static inline void inode_detach_wb(struct inode *inode)
241 WARN_ON_ONCE(!(inode->i_state & I_CLEAR));
248 * wbc_attach_fdatawrite_inode - associate wbc and inode for fdatawrite
249 * @wbc: writeback_control of interest
250 * @inode: target inode
252 * This function is to be used by __filemap_fdatawrite_range(), which is an
253 * alternative entry point into writeback code, and first ensures @inode is
254 * associated with a bdi_writeback and attaches it to @wbc.
256 static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
259 spin_lock(&inode->i_lock);
260 inode_attach_wb(inode, NULL);
261 wbc_attach_and_unlock_inode(wbc, inode);
265 * wbc_init_bio - writeback specific initializtion of bio
266 * @wbc: writeback_control for the writeback in progress
267 * @bio: bio to be initialized
269 * @bio is a part of the writeback in progress controlled by @wbc. Perform
270 * writeback specific initialization. This is used to apply the cgroup
273 static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
276 * pageout() path doesn't attach @wbc to the inode being written
277 * out. This is intentional as we don't want the function to block
278 * behind a slow cgroup. Ultimately, we want pageout() to kick off
279 * regular writeback instead of writing things out itself.
282 bio_associate_blkcg(bio, wbc->wb->blkcg_css);
285 #else /* CONFIG_CGROUP_WRITEBACK */
287 static inline void inode_attach_wb(struct inode *inode, struct page *page)
291 static inline void inode_detach_wb(struct inode *inode)
295 static inline void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
297 __releases(&inode->i_lock)
299 spin_unlock(&inode->i_lock);
302 static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
307 static inline void wbc_detach_inode(struct writeback_control *wbc)
311 static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
315 static inline void wbc_account_io(struct writeback_control *wbc,
316 struct page *page, size_t bytes)
320 static inline void cgroup_writeback_umount(void)
324 #endif /* CONFIG_CGROUP_WRITEBACK */
327 * mm/page-writeback.c
330 void laptop_io_completion(struct backing_dev_info *info);
331 void laptop_sync_completion(void);
332 void laptop_mode_sync(struct work_struct *work);
333 void laptop_mode_timer_fn(unsigned long data);
335 static inline void laptop_sync_completion(void) { }
337 bool node_dirty_ok(struct pglist_data *pgdat);
338 int wb_domain_init(struct wb_domain *dom, gfp_t gfp);
339 #ifdef CONFIG_CGROUP_WRITEBACK
340 void wb_domain_exit(struct wb_domain *dom);
343 extern struct wb_domain global_wb_domain;
345 /* These are exported to sysctl. */
346 extern int dirty_background_ratio;
347 extern unsigned long dirty_background_bytes;
348 extern int vm_dirty_ratio;
349 extern unsigned long vm_dirty_bytes;
350 extern unsigned int dirty_writeback_interval;
351 extern unsigned int dirty_expire_interval;
352 extern unsigned int dirtytime_expire_interval;
353 extern int vm_highmem_is_dirtyable;
354 extern int block_dump;
355 extern int laptop_mode;
357 extern int dirty_background_ratio_handler(struct ctl_table *table, int write,
358 void __user *buffer, size_t *lenp,
360 extern int dirty_background_bytes_handler(struct ctl_table *table, int write,
361 void __user *buffer, size_t *lenp,
363 extern int dirty_ratio_handler(struct ctl_table *table, int write,
364 void __user *buffer, size_t *lenp,
366 extern int dirty_bytes_handler(struct ctl_table *table, int write,
367 void __user *buffer, size_t *lenp,
369 int dirtytime_interval_handler(struct ctl_table *table, int write,
370 void __user *buffer, size_t *lenp, loff_t *ppos);
373 int dirty_writeback_centisecs_handler(struct ctl_table *, int,
374 void __user *, size_t *, loff_t *);
376 void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty);
377 unsigned long wb_calc_thresh(struct bdi_writeback *wb, unsigned long thresh);
379 void wb_update_bandwidth(struct bdi_writeback *wb, unsigned long start_time);
380 void balance_dirty_pages_ratelimited(struct address_space *mapping);
381 bool wb_over_bg_thresh(struct bdi_writeback *wb);
383 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
386 int generic_writepages(struct address_space *mapping,
387 struct writeback_control *wbc);
388 void tag_pages_for_writeback(struct address_space *mapping,
389 pgoff_t start, pgoff_t end);
390 int write_cache_pages(struct address_space *mapping,
391 struct writeback_control *wbc, writepage_t writepage,
393 int do_writepages(struct address_space *mapping, struct writeback_control *wbc);
394 void writeback_set_ratelimit(void);
395 void tag_pages_for_writeback(struct address_space *mapping,
396 pgoff_t start, pgoff_t end);
398 void account_page_redirty(struct page *page);
400 void sb_mark_inode_writeback(struct inode *inode);
401 void sb_clear_inode_writeback(struct inode *inode);
403 #endif /* WRITEBACK_H */