2 * include/linux/writeback.h
7 #include <linux/sched.h>
8 #include <linux/workqueue.h>
10 #include <linux/flex_proportions.h>
11 #include <linux/backing-dev-defs.h>
12 #include <linux/blk_types.h>
16 DECLARE_PER_CPU(int, dirty_throttle_leaks);
19 * The 1/4 region under the global dirty thresh is for smooth dirty throttling:
21 * (thresh - thresh/DIRTY_FULL_SCOPE, thresh)
23 * Further beyond, all dirtier tasks will enter a loop waiting (possibly long
24 * time) for the dirty pages to drop, unless written enough pages.
26 * The global dirty threshold is normally equal to the global dirty limit,
27 * except when the system suddenly allocates a lot of anonymous memory and
28 * knocks down the global dirty threshold quickly, in which case the global
29 * dirty limit will follow down slowly to prevent livelocking all dirtier tasks.
32 #define DIRTY_FULL_SCOPE (DIRTY_SCOPE / 2)
34 struct backing_dev_info;
39 enum writeback_sync_modes {
40 WB_SYNC_NONE, /* Don't wait on anything */
41 WB_SYNC_ALL, /* Wait on every mapping */
45 * A control structure which tells the writeback code what to do. These are
46 * always on the stack, and hence need no locking. They are always initialised
47 * in a manner such that unspecified fields are set to zero.
49 struct writeback_control {
50 long nr_to_write; /* Write this many pages, and decrement
51 this for each page written */
52 long pages_skipped; /* Pages which were not written */
55 * For a_ops->writepages(): if start or end are non-zero then this is
56 * a hint that the filesystem need only write out the pages inside that
57 * byterange. The byte at `end' is included in the writeout request.
62 enum writeback_sync_modes sync_mode;
64 unsigned for_kupdate:1; /* A kupdate writeback */
65 unsigned for_background:1; /* A background writeback */
66 unsigned tagged_writepages:1; /* tag-and-write to avoid livelock */
67 unsigned for_reclaim:1; /* Invoked from the page allocator */
68 unsigned range_cyclic:1; /* range_start is cyclic */
69 unsigned for_sync:1; /* sync(2) WB_SYNC_ALL writeback */
70 #ifdef CONFIG_CGROUP_WRITEBACK
71 struct bdi_writeback *wb; /* wb this writeback is issued under */
72 struct inode *inode; /* inode being written out */
74 /* foreign inode detection, see wbc_detach_inode() */
75 int wb_id; /* current wb id */
76 int wb_lcand_id; /* last foreign candidate wb id */
77 int wb_tcand_id; /* this foreign candidate wb id */
78 size_t wb_bytes; /* bytes written by current wb */
79 size_t wb_lcand_bytes; /* bytes written by last candidate */
80 size_t wb_tcand_bytes; /* bytes written by this candidate */
84 static inline int wbc_to_write_flags(struct writeback_control *wbc)
86 if (wbc->sync_mode == WB_SYNC_ALL)
88 else if (wbc->for_kupdate || wbc->for_background)
89 return REQ_BACKGROUND;
95 * A wb_domain represents a domain that wb's (bdi_writeback's) belong to
96 * and are measured against each other in. There always is one global
97 * domain, global_wb_domain, that every wb in the system is a member of.
98 * This allows measuring the relative bandwidth of each wb to distribute
99 * dirtyable memory accordingly.
105 * Scale the writeback cache size proportional to the relative
108 * We do this by keeping a floating proportion between BDIs, based
109 * on page writeback completions [end_page_writeback()]. Those
110 * devices that write out pages fastest will get the larger share,
111 * while the slower will get a smaller share.
113 * We use page writeout completions because we are interested in
114 * getting rid of dirty pages. Having them written out is the
117 * We introduce a concept of time, a period over which we measure
118 * these events, because demand can/will vary over time. The length
119 * of this period itself is measured in page writeback completions.
121 struct fprop_global completions;
122 struct timer_list period_timer; /* timer for aging of completions */
123 unsigned long period_time;
126 * The dirtyable memory and dirty threshold could be suddenly
127 * knocked down by a large amount (eg. on the startup of KVM in a
128 * swapless system). This may throw the system into deep dirty
129 * exceeded state and throttle heavy/light dirtiers alike. To
130 * retain good responsiveness, maintain global_dirty_limit for
131 * tracking slowly down to the knocked down dirty threshold.
133 * Both fields are protected by ->lock.
135 unsigned long dirty_limit_tstamp;
136 unsigned long dirty_limit;
140 * wb_domain_size_changed - memory available to a wb_domain has changed
141 * @dom: wb_domain of interest
143 * This function should be called when the amount of memory available to
144 * @dom has changed. It resets @dom's dirty limit parameters to prevent
145 * the past values which don't match the current configuration from skewing
146 * dirty throttling. Without this, when memory size of a wb_domain is
147 * greatly reduced, the dirty throttling logic may allow too many pages to
148 * be dirtied leading to consecutive unnecessary OOMs and may get stuck in
151 static inline void wb_domain_size_changed(struct wb_domain *dom)
153 spin_lock(&dom->lock);
154 dom->dirty_limit_tstamp = jiffies;
155 dom->dirty_limit = 0;
156 spin_unlock(&dom->lock);
162 struct bdi_writeback;
163 void writeback_inodes_sb(struct super_block *, enum wb_reason reason);
164 void writeback_inodes_sb_nr(struct super_block *, unsigned long nr,
165 enum wb_reason reason);
166 void try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason);
167 void sync_inodes_sb(struct super_block *);
168 void wakeup_flusher_threads(enum wb_reason reason);
169 void wakeup_flusher_threads_bdi(struct backing_dev_info *bdi,
170 enum wb_reason reason);
171 void inode_wait_for_writeback(struct inode *inode);
173 /* writeback.h requires fs.h; it, too, is not included from here. */
174 static inline void wait_on_inode(struct inode *inode)
177 wait_on_bit(&inode->i_state, __I_NEW, TASK_UNINTERRUPTIBLE);
180 #ifdef CONFIG_CGROUP_WRITEBACK
182 #include <linux/cgroup.h>
183 #include <linux/bio.h>
185 void __inode_attach_wb(struct inode *inode, struct page *page);
186 void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
188 __releases(&inode->i_lock);
189 void wbc_detach_inode(struct writeback_control *wbc);
190 void wbc_account_io(struct writeback_control *wbc, struct page *page,
192 void cgroup_writeback_umount(void);
195 * inode_attach_wb - associate an inode with its wb
196 * @inode: inode of interest
197 * @page: page being dirtied (may be NULL)
199 * If @inode doesn't have its wb, associate it with the wb matching the
200 * memcg of @page or, if @page is NULL, %current. May be called w/ or w/o
203 static inline void inode_attach_wb(struct inode *inode, struct page *page)
206 __inode_attach_wb(inode, page);
210 * inode_detach_wb - disassociate an inode from its wb
211 * @inode: inode of interest
213 * @inode is being freed. Detach from its wb.
215 static inline void inode_detach_wb(struct inode *inode)
218 WARN_ON_ONCE(!(inode->i_state & I_CLEAR));
225 * wbc_attach_fdatawrite_inode - associate wbc and inode for fdatawrite
226 * @wbc: writeback_control of interest
227 * @inode: target inode
229 * This function is to be used by __filemap_fdatawrite_range(), which is an
230 * alternative entry point into writeback code, and first ensures @inode is
231 * associated with a bdi_writeback and attaches it to @wbc.
233 static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
236 spin_lock(&inode->i_lock);
237 inode_attach_wb(inode, NULL);
238 wbc_attach_and_unlock_inode(wbc, inode);
242 * wbc_init_bio - writeback specific initializtion of bio
243 * @wbc: writeback_control for the writeback in progress
244 * @bio: bio to be initialized
246 * @bio is a part of the writeback in progress controlled by @wbc. Perform
247 * writeback specific initialization. This is used to apply the cgroup
250 static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
253 * pageout() path doesn't attach @wbc to the inode being written
254 * out. This is intentional as we don't want the function to block
255 * behind a slow cgroup. Ultimately, we want pageout() to kick off
256 * regular writeback instead of writing things out itself.
259 bio_associate_blkcg(bio, wbc->wb->blkcg_css);
262 #else /* CONFIG_CGROUP_WRITEBACK */
264 static inline void inode_attach_wb(struct inode *inode, struct page *page)
268 static inline void inode_detach_wb(struct inode *inode)
272 static inline void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
274 __releases(&inode->i_lock)
276 spin_unlock(&inode->i_lock);
279 static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
284 static inline void wbc_detach_inode(struct writeback_control *wbc)
288 static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
292 static inline void wbc_account_io(struct writeback_control *wbc,
293 struct page *page, size_t bytes)
297 static inline void cgroup_writeback_umount(void)
301 #endif /* CONFIG_CGROUP_WRITEBACK */
304 * mm/page-writeback.c
307 void laptop_io_completion(struct backing_dev_info *info);
308 void laptop_sync_completion(void);
309 void laptop_mode_sync(struct work_struct *work);
310 void laptop_mode_timer_fn(unsigned long data);
312 static inline void laptop_sync_completion(void) { }
314 bool node_dirty_ok(struct pglist_data *pgdat);
315 int wb_domain_init(struct wb_domain *dom, gfp_t gfp);
316 #ifdef CONFIG_CGROUP_WRITEBACK
317 void wb_domain_exit(struct wb_domain *dom);
320 extern struct wb_domain global_wb_domain;
322 /* These are exported to sysctl. */
323 extern int dirty_background_ratio;
324 extern unsigned long dirty_background_bytes;
325 extern int vm_dirty_ratio;
326 extern unsigned long vm_dirty_bytes;
327 extern unsigned int dirty_writeback_interval;
328 extern unsigned int dirty_expire_interval;
329 extern unsigned int dirtytime_expire_interval;
330 extern int vm_highmem_is_dirtyable;
331 extern int block_dump;
332 extern int laptop_mode;
334 extern int dirty_background_ratio_handler(struct ctl_table *table, int write,
335 void __user *buffer, size_t *lenp,
337 extern int dirty_background_bytes_handler(struct ctl_table *table, int write,
338 void __user *buffer, size_t *lenp,
340 extern int dirty_ratio_handler(struct ctl_table *table, int write,
341 void __user *buffer, size_t *lenp,
343 extern int dirty_bytes_handler(struct ctl_table *table, int write,
344 void __user *buffer, size_t *lenp,
346 int dirtytime_interval_handler(struct ctl_table *table, int write,
347 void __user *buffer, size_t *lenp, loff_t *ppos);
350 int dirty_writeback_centisecs_handler(struct ctl_table *, int,
351 void __user *, size_t *, loff_t *);
353 void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty);
354 unsigned long wb_calc_thresh(struct bdi_writeback *wb, unsigned long thresh);
356 void wb_update_bandwidth(struct bdi_writeback *wb, unsigned long start_time);
357 void balance_dirty_pages_ratelimited(struct address_space *mapping);
358 bool wb_over_bg_thresh(struct bdi_writeback *wb);
360 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
363 int generic_writepages(struct address_space *mapping,
364 struct writeback_control *wbc);
365 void tag_pages_for_writeback(struct address_space *mapping,
366 pgoff_t start, pgoff_t end);
367 int write_cache_pages(struct address_space *mapping,
368 struct writeback_control *wbc, writepage_t writepage,
370 int do_writepages(struct address_space *mapping, struct writeback_control *wbc);
371 void writeback_set_ratelimit(void);
372 void tag_pages_for_writeback(struct address_space *mapping,
373 pgoff_t start, pgoff_t end);
375 void account_page_redirty(struct page *page);
377 void sb_mark_inode_writeback(struct inode *inode);
378 void sb_clear_inode_writeback(struct inode *inode);
380 #endif /* WRITEBACK_H */