blob: e12d92808e983c4b6e129c4304ac50f307d20f6d [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001/* SPDX-License-Identifier: GPL-2.0 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
Uwe Zeisbergerf30c2262006-10-03 23:01:26 +02003 * include/linux/writeback.h
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 */
5#ifndef WRITEBACK_H
6#define WRITEBACK_H
7
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +04008#include <linux/sched.h>
Kent Overstreeta27bb332013-05-07 16:19:08 -07009#include <linux/workqueue.h>
Jens Axboef5ff8422007-09-21 09:19:54 +020010#include <linux/fs.h>
Tejun Heo380c27c2015-05-22 18:23:21 -040011#include <linux/flex_proportions.h>
Tejun Heo21c63212015-05-28 14:50:49 -040012#include <linux/backing-dev-defs.h>
Jens Axboe76372412016-11-01 10:00:38 -060013#include <linux/blk_types.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040014
Christoph Hellwig2f8b5442016-11-01 07:40:13 -060015struct bio;
16
Wu Fengguang54848d72011-04-05 13:21:19 -060017DECLARE_PER_CPU(int, dirty_throttle_leaks);
18
Wu Fengguangffd1f602011-06-19 22:18:42 -060019/*
Wu Fengguang1a12d8b2010-08-29 13:28:09 -060020 * The 1/4 region under the global dirty thresh is for smooth dirty throttling:
21 *
22 * (thresh - thresh/DIRTY_FULL_SCOPE, thresh)
23 *
Wu Fengguangffd1f602011-06-19 22:18:42 -060024 * Further beyond, all dirtier tasks will enter a loop waiting (possibly long
25 * time) for the dirty pages to drop, unless written enough pages.
26 *
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.
31 */
Wu Fengguang1a12d8b2010-08-29 13:28:09 -060032#define DIRTY_SCOPE 8
33#define DIRTY_FULL_SCOPE (DIRTY_SCOPE / 2)
Wu Fengguangffd1f602011-06-19 22:18:42 -060034
Linus Torvalds1da177e2005-04-16 15:20:36 -070035struct backing_dev_info;
36
Linus Torvalds1da177e2005-04-16 15:20:36 -070037/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070038 * fs/fs-writeback.c
39 */
40enum writeback_sync_modes {
41 WB_SYNC_NONE, /* Don't wait on anything */
42 WB_SYNC_ALL, /* Wait on every mapping */
Linus Torvalds1da177e2005-04-16 15:20:36 -070043};
44
45/*
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -060046 * why some writeback work was initiated
47 */
48enum wb_reason {
49 WB_REASON_BACKGROUND,
Johannes Weiner726d0612017-02-24 14:56:14 -080050 WB_REASON_VMSCAN,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -060051 WB_REASON_SYNC,
52 WB_REASON_PERIODIC,
53 WB_REASON_LAPTOP_TIMER,
54 WB_REASON_FREE_MORE_MEM,
55 WB_REASON_FS_FREE_SPACE,
Wanpeng Lifc6df802013-07-08 16:00:15 -070056 /*
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.
61 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -060062 WB_REASON_FORKER_THREAD,
63
64 WB_REASON_MAX,
65};
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -060066
67/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 * 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.
71 */
72struct writeback_control {
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 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 */
76
77 /*
Andrew Morton95468fd2012-03-05 15:06:02 -080078 * For a_ops->writepages(): if start or end are non-zero then this is
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 * 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.
81 */
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -070082 loff_t range_start;
83 loff_t range_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Richard Kennedy4cd90692012-04-25 14:53:05 +010085 enum writeback_sync_modes sync_mode;
86
Andrew Morton22905f72005-11-16 15:07:01 -080087 unsigned for_kupdate:1; /* A kupdate writeback */
Wu Fengguangb17621f2009-12-03 13:54:25 +010088 unsigned for_background:1; /* A background writeback */
Wu Fengguang6e6938b2010-06-06 10:38:15 -060089 unsigned tagged_writepages:1; /* tag-and-write to avoid livelock */
Andrew Morton22905f72005-11-16 15:07:01 -080090 unsigned for_reclaim:1; /* Invoked from the page allocator */
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -070091 unsigned range_cyclic:1; /* range_start is cyclic */
Dave Chinner7747bd42013-07-02 22:38:35 +100092 unsigned for_sync:1; /* sync(2) WB_SYNC_ALL writeback */
Tejun Heob16b1de2015-06-02 08:39:48 -060093#ifdef CONFIG_CGROUP_WRITEBACK
94 struct bdi_writeback *wb; /* wb this writeback is issued under */
Tejun Heo2a814902015-05-28 14:50:51 -040095 struct inode *inode; /* inode being written out */
96
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 */
Tejun Heob16b1de2015-06-02 08:39:48 -0600104#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105};
106
Jens Axboe76372412016-11-01 10:00:38 -0600107static inline int wbc_to_write_flags(struct writeback_control *wbc)
108{
109 if (wbc->sync_mode == WB_SYNC_ALL)
110 return REQ_SYNC;
Jens Axboe13edd5e2016-11-01 10:01:35 -0600111 else if (wbc->for_kupdate || wbc->for_background)
112 return REQ_BACKGROUND;
Jens Axboe76372412016-11-01 10:00:38 -0600113
114 return 0;
115}
116
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117/*
Tejun Heo380c27c2015-05-22 18:23:21 -0400118 * 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.
123 */
124struct wb_domain {
Tejun Heodcc25ae2015-05-22 18:23:22 -0400125 spinlock_t lock;
126
Tejun Heo380c27c2015-05-22 18:23:21 -0400127 /*
128 * Scale the writeback cache size proportional to the relative
129 * writeout speed.
130 *
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.
135 *
136 * We use page writeout completions because we are interested in
137 * getting rid of dirty pages. Having them written out is the
138 * primary goal.
139 *
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.
143 */
144 struct fprop_global completions;
145 struct timer_list period_timer; /* timer for aging of completions */
146 unsigned long period_time;
Tejun Heodcc25ae2015-05-22 18:23:22 -0400147
148 /*
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.
155 *
156 * Both fields are protected by ->lock.
157 */
158 unsigned long dirty_limit_tstamp;
159 unsigned long dirty_limit;
Tejun Heo380c27c2015-05-22 18:23:21 -0400160};
161
Tejun Heo2529bb32015-05-22 18:23:34 -0400162/**
163 * wb_domain_size_changed - memory available to a wb_domain has changed
164 * @dom: wb_domain of interest
165 *
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
172 * that situation.
173 */
174static inline void wb_domain_size_changed(struct wb_domain *dom)
175{
176 spin_lock(&dom->lock);
177 dom->dirty_limit_tstamp = jiffies;
178 dom->dirty_limit = 0;
179 spin_unlock(&dom->lock);
180}
181
Tejun Heo380c27c2015-05-22 18:23:21 -0400182/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 * fs/fs-writeback.c
184 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200185struct bdi_writeback;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600186void writeback_inodes_sb(struct super_block *, enum wb_reason reason);
187void writeback_inodes_sb_nr(struct super_block *, unsigned long nr,
188 enum wb_reason reason);
Tejun Heof30a7d02015-05-22 17:14:00 -0400189bool try_to_writeback_inodes_sb(struct super_block *, enum wb_reason reason);
190bool try_to_writeback_inodes_sb_nr(struct super_block *, unsigned long nr,
191 enum wb_reason reason);
Jan Kara0dc83bd2014-02-21 11:19:04 +0100192void sync_inodes_sb(struct super_block *);
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600193void wakeup_flusher_threads(long nr_pages, enum wb_reason reason);
Jan Kara169ebd92012-05-03 14:48:03 +0200194void inode_wait_for_writeback(struct inode *inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
196/* writeback.h requires fs.h; it, too, is not included from here. */
197static inline void wait_on_inode(struct inode *inode)
198{
199 might_sleep();
NeilBrown74316202014-07-07 15:16:04 +1000200 wait_on_bit(&inode->i_state, __I_NEW, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201}
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700202
Tejun Heo21c63212015-05-28 14:50:49 -0400203#ifdef CONFIG_CGROUP_WRITEBACK
204
Tejun Heob16b1de2015-06-02 08:39:48 -0600205#include <linux/cgroup.h>
206#include <linux/bio.h>
207
Tejun Heo21c63212015-05-28 14:50:49 -0400208void __inode_attach_wb(struct inode *inode, struct page *page);
Tejun Heob16b1de2015-06-02 08:39:48 -0600209void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
210 struct inode *inode)
211 __releases(&inode->i_lock);
212void wbc_detach_inode(struct writeback_control *wbc);
Tejun Heo2a814902015-05-28 14:50:51 -0400213void wbc_account_io(struct writeback_control *wbc, struct page *page,
214 size_t bytes);
Tejun Heoa1a0e232016-02-29 18:28:53 -0500215void cgroup_writeback_umount(void);
Tejun Heo21c63212015-05-28 14:50:49 -0400216
217/**
218 * inode_attach_wb - associate an inode with its wb
219 * @inode: inode of interest
220 * @page: page being dirtied (may be NULL)
221 *
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
224 * @inode->i_lock.
225 */
226static inline void inode_attach_wb(struct inode *inode, struct page *page)
227{
228 if (!inode->i_wb)
229 __inode_attach_wb(inode, page);
230}
231
232/**
233 * inode_detach_wb - disassociate an inode from its wb
234 * @inode: inode of interest
235 *
236 * @inode is being freed. Detach from its wb.
237 */
238static inline void inode_detach_wb(struct inode *inode)
239{
240 if (inode->i_wb) {
Jan Karaf7597412017-03-23 01:37:00 +0100241 WARN_ON_ONCE(!(inode->i_state & I_CLEAR));
Tejun Heo21c63212015-05-28 14:50:49 -0400242 wb_put(inode->i_wb);
243 inode->i_wb = NULL;
244 }
245}
246
Tejun Heob16b1de2015-06-02 08:39:48 -0600247/**
248 * wbc_attach_fdatawrite_inode - associate wbc and inode for fdatawrite
249 * @wbc: writeback_control of interest
250 * @inode: target inode
251 *
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.
255 */
256static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
257 struct inode *inode)
258{
259 spin_lock(&inode->i_lock);
260 inode_attach_wb(inode, NULL);
261 wbc_attach_and_unlock_inode(wbc, inode);
262}
263
264/**
265 * wbc_init_bio - writeback specific initializtion of bio
266 * @wbc: writeback_control for the writeback in progress
267 * @bio: bio to be initialized
268 *
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
271 * writeback context.
272 */
273static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
274{
275 /*
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.
280 */
281 if (wbc->wb)
282 bio_associate_blkcg(bio, wbc->wb->blkcg_css);
283}
284
Tejun Heo21c63212015-05-28 14:50:49 -0400285#else /* CONFIG_CGROUP_WRITEBACK */
286
287static inline void inode_attach_wb(struct inode *inode, struct page *page)
288{
289}
290
291static inline void inode_detach_wb(struct inode *inode)
292{
293}
294
Tejun Heob16b1de2015-06-02 08:39:48 -0600295static inline void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
296 struct inode *inode)
297 __releases(&inode->i_lock)
298{
299 spin_unlock(&inode->i_lock);
300}
301
302static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
303 struct inode *inode)
304{
305}
306
307static inline void wbc_detach_inode(struct writeback_control *wbc)
308{
309}
310
311static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
312{
313}
314
Tejun Heo2a814902015-05-28 14:50:51 -0400315static inline void wbc_account_io(struct writeback_control *wbc,
316 struct page *page, size_t bytes)
317{
318}
319
Tejun Heoa1a0e232016-02-29 18:28:53 -0500320static inline void cgroup_writeback_umount(void)
321{
322}
323
Tejun Heo21c63212015-05-28 14:50:49 -0400324#endif /* CONFIG_CGROUP_WRITEBACK */
325
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326/*
327 * mm/page-writeback.c
328 */
Jens Axboec2c49862010-05-20 09:18:47 +0200329#ifdef CONFIG_BLOCK
Matthew Garrett31373d02010-04-06 14:25:14 +0200330void laptop_io_completion(struct backing_dev_info *info);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331void laptop_sync_completion(void);
Matthew Garrett31373d02010-04-06 14:25:14 +0200332void laptop_mode_sync(struct work_struct *work);
333void laptop_mode_timer_fn(unsigned long data);
Jens Axboec2c49862010-05-20 09:18:47 +0200334#else
335static inline void laptop_sync_completion(void) { }
336#endif
Mel Gorman281e3722016-07-28 15:46:11 -0700337bool node_dirty_ok(struct pglist_data *pgdat);
Tejun Heo380c27c2015-05-22 18:23:21 -0400338int wb_domain_init(struct wb_domain *dom, gfp_t gfp);
Tejun Heo841710a2015-05-22 18:23:33 -0400339#ifdef CONFIG_CGROUP_WRITEBACK
340void wb_domain_exit(struct wb_domain *dom);
341#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342
Tejun Heodcc25ae2015-05-22 18:23:22 -0400343extern struct wb_domain global_wb_domain;
Wu Fengguangc42843f2011-03-02 15:54:09 -0600344
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345/* These are exported to sysctl. */
346extern int dirty_background_ratio;
David Rientjes2da02992009-01-06 14:39:31 -0800347extern unsigned long dirty_background_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348extern int vm_dirty_ratio;
David Rientjes2da02992009-01-06 14:39:31 -0800349extern unsigned long vm_dirty_bytes;
Alexey Dobriyan704503d2009-03-31 15:23:18 -0700350extern unsigned int dirty_writeback_interval;
351extern unsigned int dirty_expire_interval;
Theodore Ts'o1efff912015-03-17 12:23:32 -0400352extern unsigned int dirtytime_expire_interval;
Bron Gondwana195cf4532008-02-04 22:29:20 -0800353extern int vm_highmem_is_dirtyable;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354extern int block_dump;
355extern int laptop_mode;
356
David Rientjes2da02992009-01-06 14:39:31 -0800357extern int dirty_background_ratio_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700358 void __user *buffer, size_t *lenp,
David Rientjes2da02992009-01-06 14:39:31 -0800359 loff_t *ppos);
360extern int dirty_background_bytes_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700361 void __user *buffer, size_t *lenp,
David Rientjes2da02992009-01-06 14:39:31 -0800362 loff_t *ppos);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700363extern int dirty_ratio_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700364 void __user *buffer, size_t *lenp,
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700365 loff_t *ppos);
David Rientjes2da02992009-01-06 14:39:31 -0800366extern int dirty_bytes_handler(struct ctl_table *table, int write,
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700367 void __user *buffer, size_t *lenp,
David Rientjes2da02992009-01-06 14:39:31 -0800368 loff_t *ppos);
Theodore Ts'o1efff912015-03-17 12:23:32 -0400369int dirtytime_interval_handler(struct ctl_table *table, int write,
370 void __user *buffer, size_t *lenp, loff_t *ppos);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700371
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372struct ctl_table;
Alexey Dobriyan8d65af72009-09-23 15:57:19 -0700373int dirty_writeback_centisecs_handler(struct ctl_table *, int,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 void __user *, size_t *, loff_t *);
375
Wu Fengguang16c40422010-08-11 14:17:39 -0700376void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty);
Tejun Heo0d960a32015-05-22 18:23:19 -0400377unsigned long wb_calc_thresh(struct bdi_writeback *wb, unsigned long thresh);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700378
Tejun Heo8a731792015-05-22 18:23:20 -0400379void wb_update_bandwidth(struct bdi_writeback *wb, unsigned long start_time);
Namjae Jeond0e1d662012-12-11 16:00:21 -0800380void balance_dirty_pages_ratelimited(struct address_space *mapping);
Tejun Heoaa661bb2015-05-22 18:23:31 -0400381bool wb_over_bg_thresh(struct bdi_writeback *wb);
Andrew Mortonfa5a7342006-03-24 03:18:10 -0800382
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700383typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
384 void *data);
385
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700386int generic_writepages(struct address_space *mapping,
387 struct writeback_control *wbc);
Eric Sandeen5b41d922010-10-27 21:30:13 -0400388void tag_pages_for_writeback(struct address_space *mapping,
389 pgoff_t start, pgoff_t end);
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700390int write_cache_pages(struct address_space *mapping,
391 struct writeback_control *wbc, writepage_t writepage,
392 void *data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393int do_writepages(struct address_space *mapping, struct writeback_control *wbc);
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -0700394void writeback_set_ratelimit(void);
Namhyung Kim92c09c02010-10-26 14:22:03 -0700395void tag_pages_for_writeback(struct address_space *mapping,
396 pgoff_t start, pgoff_t end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
Wu Fengguang2f800fb2011-08-08 15:22:00 -0600398void account_page_redirty(struct page *page);
399
Dave Chinner6c60d2b2016-07-26 15:21:50 -0700400void sb_mark_inode_writeback(struct inode *inode);
401void sb_clear_inode_writeback(struct inode *inode);
402
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403#endif /* WRITEBACK_H */