1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * include/linux/writeback.h
4 */
5#ifndef WRITEBACK_H
6#define WRITEBACK_H
7
8#include <linux/sched.h>
9#include <linux/workqueue.h>
10#include <linux/fs.h>
11#include <linux/flex_proportions.h>
12#include <linux/backing-dev-defs.h>
13#include <linux/blk_types.h>
14#include <linux/pagevec.h>
15
16struct bio;
17
18DECLARE_PER_CPU(int, dirty_throttle_leaks);
19
20/*
21 * The global dirty threshold is normally equal to the global dirty limit,
22 * except when the system suddenly allocates a lot of anonymous memory and
23 * knocks down the global dirty threshold quickly, in which case the global
24 * dirty limit will follow down slowly to prevent livelocking all dirtier tasks.
25 */
26#define DIRTY_SCOPE		8
27
28struct backing_dev_info;
29
30/*
31 * fs/fs-writeback.c
32 */
33enum writeback_sync_modes {
34	WB_SYNC_NONE,	/* Don't wait on anything */
35	WB_SYNC_ALL,	/* Wait on every mapping */
36};
37
38/*
39 * A control structure which tells the writeback code what to do.  These are
40 * always on the stack, and hence need no locking.  They are always initialised
41 * in a manner such that unspecified fields are set to zero.
42 */
43struct writeback_control {
44	/* public fields that can be set and/or consumed by the caller: */
45	long nr_to_write;		/* Write this many pages, and decrement
46					   this for each page written */
47	long pages_skipped;		/* Pages which were not written */
48
49	/*
50	 * For a_ops->writepages(): if start or end are non-zero then this is
51	 * a hint that the filesystem need only write out the pages inside that
52	 * byterange.  The byte at `end' is included in the writeout request.
53	 */
54	loff_t range_start;
55	loff_t range_end;
56
57	enum writeback_sync_modes sync_mode;
58
59	unsigned for_kupdate:1;		/* A kupdate writeback */
60	unsigned for_background:1;	/* A background writeback */
61	unsigned tagged_writepages:1;	/* tag-and-write to avoid livelock */
62	unsigned for_reclaim:1;		/* Invoked from the page allocator */
63	unsigned range_cyclic:1;	/* range_start is cyclic */
64	unsigned for_sync:1;		/* sync(2) WB_SYNC_ALL writeback */
65	unsigned unpinned_netfs_wb:1;	/* Cleared I_PINNING_NETFS_WB */
66
67	/*
68	 * When writeback IOs are bounced through async layers, only the
69	 * initial synchronous phase should be accounted towards inode
70	 * cgroup ownership arbitration to avoid confusion.  Later stages
71	 * can set the following flag to disable the accounting.
72	 */
73	unsigned no_cgroup_owner:1;
74
75	/* To enable batching of swap writes to non-block-device backends,
76	 * "plug" can be set point to a 'struct swap_iocb *'.  When all swap
77	 * writes have been submitted, if with swap_iocb is not NULL,
78	 * swap_write_unplug() should be called.
79	 */
80	struct swap_iocb **swap_plug;
81
82	/* internal fields used by the ->writepages implementation: */
83	struct folio_batch fbatch;
84	pgoff_t index;
85	int saved_err;
86
87#ifdef CONFIG_CGROUP_WRITEBACK
88	struct bdi_writeback *wb;	/* wb this writeback is issued under */
89	struct inode *inode;		/* inode being written out */
90
91	/* foreign inode detection, see wbc_detach_inode() */
92	int wb_id;			/* current wb id */
93	int wb_lcand_id;		/* last foreign candidate wb id */
94	int wb_tcand_id;		/* this foreign candidate wb id */
95	size_t wb_bytes;		/* bytes written by current wb */
96	size_t wb_lcand_bytes;		/* bytes written by last candidate */
97	size_t wb_tcand_bytes;		/* bytes written by this candidate */
98#endif
99};
100
101static inline blk_opf_t wbc_to_write_flags(struct writeback_control *wbc)
102{
103	blk_opf_t flags = 0;
104
105	if (wbc->sync_mode == WB_SYNC_ALL)
106		flags |= REQ_SYNC;
107	else if (wbc->for_kupdate || wbc->for_background)
108		flags |= REQ_BACKGROUND;
109
110	return flags;
111}
112
113#ifdef CONFIG_CGROUP_WRITEBACK
114#define wbc_blkcg_css(wbc) \
115	((wbc)->wb ? (wbc)->wb->blkcg_css : blkcg_root_css)
116#else
117#define wbc_blkcg_css(wbc)		(blkcg_root_css)
118#endif /* CONFIG_CGROUP_WRITEBACK */
119
120/*
121 * A wb_domain represents a domain that wb's (bdi_writeback's) belong to
122 * and are measured against each other in.  There always is one global
123 * domain, global_wb_domain, that every wb in the system is a member of.
124 * This allows measuring the relative bandwidth of each wb to distribute
125 * dirtyable memory accordingly.
126 */
127struct wb_domain {
128	spinlock_t lock;
129
130	/*
131	 * Scale the writeback cache size proportional to the relative
132	 * writeout speed.
133	 *
134	 * We do this by keeping a floating proportion between BDIs, based
135	 * on page writeback completions [end_page_writeback()]. Those
136	 * devices that write out pages fastest will get the larger share,
137	 * while the slower will get a smaller share.
138	 *
139	 * We use page writeout completions because we are interested in
140	 * getting rid of dirty pages. Having them written out is the
141	 * primary goal.
142	 *
143	 * We introduce a concept of time, a period over which we measure
144	 * these events, because demand can/will vary over time. The length
145	 * of this period itself is measured in page writeback completions.
146	 */
147	struct fprop_global completions;
148	struct timer_list period_timer;	/* timer for aging of completions */
149	unsigned long period_time;
150
151	/*
152	 * The dirtyable memory and dirty threshold could be suddenly
153	 * knocked down by a large amount (eg. on the startup of KVM in a
154	 * swapless system). This may throw the system into deep dirty
155	 * exceeded state and throttle heavy/light dirtiers alike. To
156	 * retain good responsiveness, maintain global_dirty_limit for
157	 * tracking slowly down to the knocked down dirty threshold.
158	 *
159	 * Both fields are protected by ->lock.
160	 */
161	unsigned long dirty_limit_tstamp;
162	unsigned long dirty_limit;
163};
164
165/**
166 * wb_domain_size_changed - memory available to a wb_domain has changed
167 * @dom: wb_domain of interest
168 *
169 * This function should be called when the amount of memory available to
170 * @dom has changed.  It resets @dom's dirty limit parameters to prevent
171 * the past values which don't match the current configuration from skewing
172 * dirty throttling.  Without this, when memory size of a wb_domain is
173 * greatly reduced, the dirty throttling logic may allow too many pages to
174 * be dirtied leading to consecutive unnecessary OOMs and may get stuck in
175 * that situation.
176 */
177static inline void wb_domain_size_changed(struct wb_domain *dom)
178{
179	spin_lock(&dom->lock);
180	dom->dirty_limit_tstamp = jiffies;
181	dom->dirty_limit = 0;
182	spin_unlock(&dom->lock);
183}
184
185/*
186 * fs/fs-writeback.c
187 */
188struct bdi_writeback;
189void writeback_inodes_sb(struct super_block *, enum wb_reason reason);
190void writeback_inodes_sb_nr(struct super_block *, unsigned long nr,
191							enum wb_reason reason);
192void try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason);
193void sync_inodes_sb(struct super_block *);
194void wakeup_flusher_threads(enum wb_reason reason);
195void wakeup_flusher_threads_bdi(struct backing_dev_info *bdi,
196				enum wb_reason reason);
197void inode_wait_for_writeback(struct inode *inode);
198void inode_io_list_del(struct inode *inode);
199
200/* writeback.h requires fs.h; it, too, is not included from here. */
201static inline void wait_on_inode(struct inode *inode)
202{
203	wait_on_bit(&inode->i_state, __I_NEW, TASK_UNINTERRUPTIBLE);
204}
205
206#ifdef CONFIG_CGROUP_WRITEBACK
207
208#include <linux/cgroup.h>
209#include <linux/bio.h>
210
211void __inode_attach_wb(struct inode *inode, struct folio *folio);
212void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
213				 struct inode *inode)
214	__releases(&inode->i_lock);
215void wbc_detach_inode(struct writeback_control *wbc);
216void wbc_account_cgroup_owner(struct writeback_control *wbc, struct page *page,
217			      size_t bytes);
218int cgroup_writeback_by_id(u64 bdi_id, int memcg_id,
219			   enum wb_reason reason, struct wb_completion *done);
220void cgroup_writeback_umount(void);
221bool cleanup_offline_cgwb(struct bdi_writeback *wb);
222
223/**
224 * inode_attach_wb - associate an inode with its wb
225 * @inode: inode of interest
226 * @folio: folio being dirtied (may be NULL)
227 *
228 * If @inode doesn't have its wb, associate it with the wb matching the
229 * memcg of @folio or, if @folio is NULL, %current.  May be called w/ or w/o
230 * @inode->i_lock.
231 */
232static inline void inode_attach_wb(struct inode *inode, struct folio *folio)
233{
234	if (!inode->i_wb)
235		__inode_attach_wb(inode, folio);
236}
237
238/**
239 * inode_detach_wb - disassociate an inode from its wb
240 * @inode: inode of interest
241 *
242 * @inode is being freed.  Detach from its wb.
243 */
244static inline void inode_detach_wb(struct inode *inode)
245{
246	if (inode->i_wb) {
247		WARN_ON_ONCE(!(inode->i_state & I_CLEAR));
248		wb_put(inode->i_wb);
249		inode->i_wb = NULL;
250	}
251}
252
253/**
254 * wbc_attach_fdatawrite_inode - associate wbc and inode for fdatawrite
255 * @wbc: writeback_control of interest
256 * @inode: target inode
257 *
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.
261 */
262static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
263					       struct inode *inode)
264{
265	spin_lock(&inode->i_lock);
266	inode_attach_wb(inode, NULL);
267	wbc_attach_and_unlock_inode(wbc, inode);
268}
269
270/**
271 * wbc_init_bio - writeback specific initializtion of bio
272 * @wbc: writeback_control for the writeback in progress
273 * @bio: bio to be initialized
274 *
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
278 * a device.
279 */
280static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
281{
282	/*
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.
287	 */
288	if (wbc->wb)
289		bio_associate_blkg_from_css(bio, wbc->wb->blkcg_css);
290}
291
292#else	/* CONFIG_CGROUP_WRITEBACK */
293
294static inline void inode_attach_wb(struct inode *inode, struct folio *folio)
295{
296}
297
298static inline void inode_detach_wb(struct inode *inode)
299{
300}
301
302static inline void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
303					       struct inode *inode)
304	__releases(&inode->i_lock)
305{
306	spin_unlock(&inode->i_lock);
307}
308
309static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
310					       struct inode *inode)
311{
312}
313
314static inline void wbc_detach_inode(struct writeback_control *wbc)
315{
316}
317
318static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
319{
320}
321
322static inline void wbc_account_cgroup_owner(struct writeback_control *wbc,
323					    struct page *page, size_t bytes)
324{
325}
326
327static inline void cgroup_writeback_umount(void)
328{
329}
330
331#endif	/* CONFIG_CGROUP_WRITEBACK */
332
333/*
334 * mm/page-writeback.c
335 */
336void laptop_io_completion(struct backing_dev_info *info);
337void laptop_sync_completion(void);
338void laptop_mode_timer_fn(struct timer_list *t);
339bool node_dirty_ok(struct pglist_data *pgdat);
340int wb_domain_init(struct wb_domain *dom, gfp_t gfp);
341#ifdef CONFIG_CGROUP_WRITEBACK
342void wb_domain_exit(struct wb_domain *dom);
343#endif
344
345extern struct wb_domain global_wb_domain;
346
347/* These are exported to sysctl. */
348extern unsigned int dirty_writeback_interval;
349extern unsigned int dirty_expire_interval;
350extern unsigned int dirtytime_expire_interval;
351extern int laptop_mode;
352
353int dirtytime_interval_handler(struct ctl_table *table, int write,
354		void *buffer, size_t *lenp, loff_t *ppos);
355
356void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty);
357unsigned long wb_calc_thresh(struct bdi_writeback *wb, unsigned long thresh);
358
359void wb_update_bandwidth(struct bdi_writeback *wb);
360
361/* Invoke balance dirty pages in async mode. */
362#define BDP_ASYNC 0x0001
363
364void balance_dirty_pages_ratelimited(struct address_space *mapping);
365int balance_dirty_pages_ratelimited_flags(struct address_space *mapping,
366		unsigned int flags);
367
368bool wb_over_bg_thresh(struct bdi_writeback *wb);
369
370struct folio *writeback_iter(struct address_space *mapping,
371		struct writeback_control *wbc, struct folio *folio, int *error);
372
373typedef int (*writepage_t)(struct folio *folio, struct writeback_control *wbc,
374				void *data);
375
376int write_cache_pages(struct address_space *mapping,
377		      struct writeback_control *wbc, writepage_t writepage,
378		      void *data);
379int do_writepages(struct address_space *mapping, struct writeback_control *wbc);
380void writeback_set_ratelimit(void);
381void tag_pages_for_writeback(struct address_space *mapping,
382			     pgoff_t start, pgoff_t end);
383
384bool filemap_dirty_folio(struct address_space *mapping, struct folio *folio);
385bool folio_redirty_for_writepage(struct writeback_control *, struct folio *);
386bool redirty_page_for_writepage(struct writeback_control *, struct page *);
387
388void sb_mark_inode_writeback(struct inode *inode);
389void sb_clear_inode_writeback(struct inode *inode);
390
391#endif		/* WRITEBACK_H */
392