writeback.h

  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 range_cyclic:1;	/* range_start is cyclic */
 63	unsigned for_sync:1;		/* sync(2) WB_SYNC_ALL writeback */
 64	unsigned unpinned_netfs_wb:1;	/* Cleared I_PINNING_NETFS_WB */
 65
 66	/*
 67	 * When writeback IOs are bounced through async layers, only the
 68	 * initial synchronous phase should be accounted towards inode
 69	 * cgroup ownership arbitration to avoid confusion.  Later stages
 70	 * can set the following flag to disable the accounting.
 71	 */
 72	unsigned no_cgroup_owner:1;
 73
 74	/* internal fields used by the ->writepages implementation: */
 75	struct folio_batch fbatch;
 76	pgoff_t index;
 77	int saved_err;
 78
 79#ifdef CONFIG_CGROUP_WRITEBACK
 80	struct bdi_writeback *wb;	/* wb this writeback is issued under */
 81	struct inode *inode;		/* inode being written out */
 82
 83	/* foreign inode detection, see wbc_detach_inode() */
 84	int wb_id;			/* current wb id */
 85	int wb_lcand_id;		/* last foreign candidate wb id */
 86	int wb_tcand_id;		/* this foreign candidate wb id */
 87	size_t wb_bytes;		/* bytes written by current wb */
 88	size_t wb_lcand_bytes;		/* bytes written by last candidate */
 89	size_t wb_tcand_bytes;		/* bytes written by this candidate */
 90#endif
 91};
 92
 93static inline blk_opf_t wbc_to_write_flags(struct writeback_control *wbc)
 94{
 95	blk_opf_t flags = 0;
 96
 97	if (wbc->sync_mode == WB_SYNC_ALL)
 98		flags |= REQ_SYNC;
 99	else if (wbc->for_kupdate || wbc->for_background)
100		flags |= REQ_BACKGROUND;
101
102	return flags;
103}
104
105#ifdef CONFIG_CGROUP_WRITEBACK
106#define wbc_blkcg_css(wbc) \
107	((wbc)->wb ? (wbc)->wb->blkcg_css : blkcg_root_css)
108#else
109#define wbc_blkcg_css(wbc)		(blkcg_root_css)
110#endif /* CONFIG_CGROUP_WRITEBACK */
111
112/*
113 * A wb_domain represents a domain that wb's (bdi_writeback's) belong to
114 * and are measured against each other in.  There always is one global
115 * domain, global_wb_domain, that every wb in the system is a member of.
116 * This allows measuring the relative bandwidth of each wb to distribute
117 * dirtyable memory accordingly.
118 */
119struct wb_domain {
120	spinlock_t lock;
121
122	/*
123	 * Scale the writeback cache size proportional to the relative
124	 * writeout speed.
125	 *
126	 * We do this by keeping a floating proportion between BDIs, based
127	 * on page writeback completions [end_page_writeback()]. Those
128	 * devices that write out pages fastest will get the larger share,
129	 * while the slower will get a smaller share.
130	 *
131	 * We use page writeout completions because we are interested in
132	 * getting rid of dirty pages. Having them written out is the
133	 * primary goal.
134	 *
135	 * We introduce a concept of time, a period over which we measure
136	 * these events, because demand can/will vary over time. The length
137	 * of this period itself is measured in page writeback completions.
138	 */
139	struct fprop_global completions;
140	struct timer_list period_timer;	/* timer for aging of completions */
141	unsigned long period_time;
142
143	/*
144	 * The dirtyable memory and dirty threshold could be suddenly
145	 * knocked down by a large amount (eg. on the startup of KVM in a
146	 * swapless system). This may throw the system into deep dirty
147	 * exceeded state and throttle heavy/light dirtiers alike. To
148	 * retain good responsiveness, maintain global_dirty_limit for
149	 * tracking slowly down to the knocked down dirty threshold.
150	 *
151	 * Both fields are protected by ->lock.
152	 */
153	unsigned long dirty_limit_tstamp;
154	unsigned long dirty_limit;
155};
156
157/**
158 * wb_domain_size_changed - memory available to a wb_domain has changed
159 * @dom: wb_domain of interest
160 *
161 * This function should be called when the amount of memory available to
162 * @dom has changed.  It resets @dom's dirty limit parameters to prevent
163 * the past values which don't match the current configuration from skewing
164 * dirty throttling.  Without this, when memory size of a wb_domain is
165 * greatly reduced, the dirty throttling logic may allow too many pages to
166 * be dirtied leading to consecutive unnecessary OOMs and may get stuck in
167 * that situation.
168 */
169static inline void wb_domain_size_changed(struct wb_domain *dom)
170{
171	spin_lock(&dom->lock);
172	dom->dirty_limit_tstamp = jiffies;
173	dom->dirty_limit = 0;
174	spin_unlock(&dom->lock);
175}
176
177/*
178 * fs/fs-writeback.c
179 */	
180struct bdi_writeback;
181void writeback_inodes_sb(struct super_block *, enum wb_reason reason);
182void writeback_inodes_sb_nr(struct super_block *, unsigned long nr,
183							enum wb_reason reason);
184void try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason);
185void sync_inodes_sb(struct super_block *);
186void wakeup_flusher_threads(enum wb_reason reason);
187void wakeup_flusher_threads_bdi(struct backing_dev_info *bdi,
188				enum wb_reason reason);
189void inode_wait_for_writeback(struct inode *inode);
190void inode_io_list_del(struct inode *inode);
191
192static inline xa_mark_t wbc_to_tag(struct writeback_control *wbc)
193{
194	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
195		return PAGECACHE_TAG_TOWRITE;
196	return PAGECACHE_TAG_DIRTY;
197}
198
199#ifdef CONFIG_CGROUP_WRITEBACK
200
201#include <linux/cgroup.h>
202#include <linux/bio.h>
203
204void __inode_attach_wb(struct inode *inode, struct folio *folio);
205void wbc_detach_inode(struct writeback_control *wbc);
206void wbc_account_cgroup_owner(struct writeback_control *wbc, struct folio *folio,
207			      size_t bytes);
208int cgroup_writeback_by_id(u64 bdi_id, int memcg_id,
209			   enum wb_reason reason, struct wb_completion *done);
210void cgroup_writeback_umount(struct super_block *sb);
211bool cleanup_offline_cgwb(struct bdi_writeback *wb);
212
213/**
214 * inode_attach_wb - associate an inode with its wb
215 * @inode: inode of interest
216 * @folio: folio being dirtied (may be NULL)
217 *
218 * If @inode doesn't have its wb, associate it with the wb matching the
219 * memcg of @folio or, if @folio is NULL, %current.  May be called w/ or w/o
220 * @inode->i_lock.
221 */
222static inline void inode_attach_wb(struct inode *inode, struct folio *folio)
223{
224	if (!inode->i_wb)
225		__inode_attach_wb(inode, folio);
226}
227
228/**
229 * inode_detach_wb - disassociate an inode from its wb
230 * @inode: inode of interest
231 *
232 * @inode is being freed.  Detach from its wb.
233 */
234static inline void inode_detach_wb(struct inode *inode)
235{
236	if (inode->i_wb) {
237		WARN_ON_ONCE(!(inode_state_read_once(inode) & I_CLEAR));
238		wb_put(inode->i_wb);
239		inode->i_wb = NULL;
240	}
241}
242
243void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
244		struct inode *inode);
245
246/**
247 * wbc_init_bio - writeback specific initializtion of bio
248 * @wbc: writeback_control for the writeback in progress
249 * @bio: bio to be initialized
250 *
251 * @bio is a part of the writeback in progress controlled by @wbc.  Perform
252 * writeback specific initialization.  This is used to apply the cgroup
253 * writeback context.  Must be called after the bio has been associated with
254 * a device.
255 */
256static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
257{
258	/*
259	 * pageout() path doesn't attach @wbc to the inode being written
260	 * out.  This is intentional as we don't want the function to block
261	 * behind a slow cgroup.  Ultimately, we want pageout() to kick off
262	 * regular writeback instead of writing things out itself.
263	 */
264	if (wbc->wb)
265		bio_associate_blkg_from_css(bio, wbc->wb->blkcg_css);
266}
267
268void inode_switch_wbs_work_fn(struct work_struct *work);
269
270#else	/* CONFIG_CGROUP_WRITEBACK */
271
272static inline void inode_attach_wb(struct inode *inode, struct folio *folio)
273{
274}
275
276static inline void inode_detach_wb(struct inode *inode)
277{
278}
279
280static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
281					       struct inode *inode)
282{
283}
284
285static inline void wbc_detach_inode(struct writeback_control *wbc)
286{
287}
288
289static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
290{
291}
292
293static inline void wbc_account_cgroup_owner(struct writeback_control *wbc,
294					    struct folio *folio, size_t bytes)
295{
296}
297
298static inline void cgroup_writeback_umount(struct super_block *sb)
299{
300}
301
302#endif	/* CONFIG_CGROUP_WRITEBACK */
303
304/*
305 * mm/page-writeback.c
306 */
307/* consolidated parameters for balance_dirty_pages() and its subroutines */
308struct dirty_throttle_control {
309#ifdef CONFIG_CGROUP_WRITEBACK
310	struct wb_domain	*dom;
311	struct dirty_throttle_control *gdtc;	/* only set in memcg dtc's */
312#endif
313	struct bdi_writeback	*wb;
314	struct fprop_local_percpu *wb_completions;
315
316	unsigned long		avail;		/* dirtyable */
317	unsigned long		dirty;		/* file_dirty + write + nfs */
318	unsigned long		thresh;		/* dirty threshold */
319	unsigned long		bg_thresh;	/* dirty background threshold */
320	unsigned long		limit;		/* hard dirty limit */
321
322	unsigned long		wb_dirty;	/* per-wb counterparts */
323	unsigned long		wb_thresh;
324	unsigned long		wb_bg_thresh;
325
326	unsigned long		pos_ratio;
327	bool			freerun;
328	bool			dirty_exceeded;
329};
330
331void laptop_io_completion(struct backing_dev_info *info);
332void laptop_sync_completion(void);
333void laptop_mode_timer_fn(struct timer_list *t);
334bool node_dirty_ok(struct pglist_data *pgdat);
335int wb_domain_init(struct wb_domain *dom, gfp_t gfp);
336#ifdef CONFIG_CGROUP_WRITEBACK
337void wb_domain_exit(struct wb_domain *dom);
338#endif
339
340extern struct wb_domain global_wb_domain;
341
342/* These are exported to sysctl. */
343extern unsigned int dirty_writeback_interval;
344extern unsigned int dirty_expire_interval;
345extern int laptop_mode;
346
347void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty);
348unsigned long wb_calc_thresh(struct bdi_writeback *wb, unsigned long thresh);
349unsigned long cgwb_calc_thresh(struct bdi_writeback *wb);
350
351void wb_update_bandwidth(struct bdi_writeback *wb);
352
353/* Invoke balance dirty pages in async mode. */
354#define BDP_ASYNC 0x0001
355
356void balance_dirty_pages_ratelimited(struct address_space *mapping);
357int balance_dirty_pages_ratelimited_flags(struct address_space *mapping,
358		unsigned int flags);
359
360bool wb_over_bg_thresh(struct bdi_writeback *wb);
361
362struct folio *writeback_iter(struct address_space *mapping,
363		struct writeback_control *wbc, struct folio *folio, int *error);
364
365int do_writepages(struct address_space *mapping, struct writeback_control *wbc);
366void writeback_set_ratelimit(void);
367void tag_pages_for_writeback(struct address_space *mapping,
368			     pgoff_t start, pgoff_t end);
369
370bool filemap_dirty_folio(struct address_space *mapping, struct folio *folio);
371bool folio_redirty_for_writepage(struct writeback_control *, struct folio *);
372bool redirty_page_for_writepage(struct writeback_control *, struct page *);
373
374void sb_mark_inode_writeback(struct inode *inode);
375void sb_clear_inode_writeback(struct inode *inode);
376
377/*
378 * 4MB minimal write chunk size
379 */
380#define MIN_WRITEBACK_PAGES	(4096UL >> (PAGE_SHIFT - 10))
381
382#endif		/* WRITEBACK_H */