projects/linux·blob v6.5-rc3 · refs

389 lines · 12.4 KiB

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
15struct bio;
16
17DECLARE_PER_CPU(int, dirty_throttle_leaks);
18
19/*
20 * The global dirty threshold is normally equal to the global dirty limit,
21 * except when the system suddenly allocates a lot of anonymous memory and
22 * knocks down the global dirty threshold quickly, in which case the global
23 * dirty limit will follow down slowly to prevent livelocking all dirtier tasks.
24 */
25#define DIRTY_SCOPE 8
26
27struct backing_dev_info;
28
29/*
30 * fs/fs-writeback.c
31 */
32enum writeback_sync_modes {
33 WB_SYNC_NONE, /* Don't wait on anything */
34 WB_SYNC_ALL, /* Wait on every mapping */
35};
36
37/*
38 * A control structure which tells the writeback code what to do. These are
39 * always on the stack, and hence need no locking. They are always initialised
40 * in a manner such that unspecified fields are set to zero.
41 */
42struct writeback_control {
43 long nr_to_write; /* Write this many pages, and decrement
44 this for each page written */
45 long pages_skipped; /* Pages which were not written */
46
47 /*
48 * For a_ops->writepages(): if start or end are non-zero then this is
49 * a hint that the filesystem need only write out the pages inside that
50 * byterange. The byte at `end' is included in the writeout request.
51 */
52 loff_t range_start;
53 loff_t range_end;
54
55 enum writeback_sync_modes sync_mode;
56
57 unsigned for_kupdate:1; /* A kupdate writeback */
58 unsigned for_background:1; /* A background writeback */
59 unsigned tagged_writepages:1; /* tag-and-write to avoid livelock */
60 unsigned for_reclaim:1; /* Invoked from the page allocator */
61 unsigned range_cyclic:1; /* range_start is cyclic */
62 unsigned for_sync:1; /* sync(2) WB_SYNC_ALL writeback */
63 unsigned unpinned_fscache_wb:1; /* Cleared I_PINNING_FSCACHE_WB */
64
65 /*
66 * When writeback IOs are bounced through async layers, only the
67 * initial synchronous phase should be accounted towards inode
68 * cgroup ownership arbitration to avoid confusion. Later stages
69 * can set the following flag to disable the accounting.
70 */
71 unsigned no_cgroup_owner:1;
72
73 /* To enable batching of swap writes to non-block-device backends,
74 * "plug" can be set point to a 'struct swap_iocb *'. When all swap
75 * writes have been submitted, if with swap_iocb is not NULL,
76 * swap_write_unplug() should be called.
77 */
78 struct swap_iocb **swap_plug;
79
80#ifdef CONFIG_CGROUP_WRITEBACK
81 struct bdi_writeback *wb; /* wb this writeback is issued under */
82 struct inode *inode; /* inode being written out */
83
84 /* foreign inode detection, see wbc_detach_inode() */
85 int wb_id; /* current wb id */
86 int wb_lcand_id; /* last foreign candidate wb id */
87 int wb_tcand_id; /* this foreign candidate wb id */
88 size_t wb_bytes; /* bytes written by current wb */
89 size_t wb_lcand_bytes; /* bytes written by last candidate */
90 size_t wb_tcand_bytes; /* bytes written by this candidate */
91#endif
92};
93
94static inline blk_opf_t wbc_to_write_flags(struct writeback_control *wbc)
95{
96 blk_opf_t flags = 0;
97
98 if (wbc->sync_mode == WB_SYNC_ALL)
99 flags |= REQ_SYNC;
100 else if (wbc->for_kupdate || wbc->for_background)
101 flags |= REQ_BACKGROUND;
102
103 return flags;
104}
105
106#ifdef CONFIG_CGROUP_WRITEBACK
107#define wbc_blkcg_css(wbc) \
108 ((wbc)->wb ? (wbc)->wb->blkcg_css : blkcg_root_css)
109#else
110#define wbc_blkcg_css(wbc) (blkcg_root_css)
111#endif /* CONFIG_CGROUP_WRITEBACK */
112
113/*
114 * A wb_domain represents a domain that wb's (bdi_writeback's) belong to
115 * and are measured against each other in. There always is one global
116 * domain, global_wb_domain, that every wb in the system is a member of.
117 * This allows measuring the relative bandwidth of each wb to distribute
118 * dirtyable memory accordingly.
119 */
120struct wb_domain {
121 spinlock_t lock;
122
123 /*
124 * Scale the writeback cache size proportional to the relative
125 * writeout speed.
126 *
127 * We do this by keeping a floating proportion between BDIs, based
128 * on page writeback completions [end_page_writeback()]. Those
129 * devices that write out pages fastest will get the larger share,
130 * while the slower will get a smaller share.
131 *
132 * We use page writeout completions because we are interested in
133 * getting rid of dirty pages. Having them written out is the
134 * primary goal.
135 *
136 * We introduce a concept of time, a period over which we measure
137 * these events, because demand can/will vary over time. The length
138 * of this period itself is measured in page writeback completions.
139 */
140 struct fprop_global completions;
141 struct timer_list period_timer; /* timer for aging of completions */
142 unsigned long period_time;
143
144 /*
145 * The dirtyable memory and dirty threshold could be suddenly
146 * knocked down by a large amount (eg. on the startup of KVM in a
147 * swapless system). This may throw the system into deep dirty
148 * exceeded state and throttle heavy/light dirtiers alike. To
149 * retain good responsiveness, maintain global_dirty_limit for
150 * tracking slowly down to the knocked down dirty threshold.
151 *
152 * Both fields are protected by ->lock.
153 */
154 unsigned long dirty_limit_tstamp;
155 unsigned long dirty_limit;
156};
157
158/**
159 * wb_domain_size_changed - memory available to a wb_domain has changed
160 * @dom: wb_domain of interest
161 *
162 * This function should be called when the amount of memory available to
163 * @dom has changed. It resets @dom's dirty limit parameters to prevent
164 * the past values which don't match the current configuration from skewing
165 * dirty throttling. Without this, when memory size of a wb_domain is
166 * greatly reduced, the dirty throttling logic may allow too many pages to
167 * be dirtied leading to consecutive unnecessary OOMs and may get stuck in
168 * that situation.
169 */
170static inline void wb_domain_size_changed(struct wb_domain *dom)
171{
172 spin_lock(&dom->lock);
173 dom->dirty_limit_tstamp = jiffies;
174 dom->dirty_limit = 0;
175 spin_unlock(&dom->lock);
176}
177
178/*
179 * fs/fs-writeback.c
180 */
181struct bdi_writeback;
182void writeback_inodes_sb(struct super_block *, enum wb_reason reason);
183void writeback_inodes_sb_nr(struct super_block *, unsigned long nr,
184 enum wb_reason reason);
185void try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason);
186void sync_inodes_sb(struct super_block *);
187void wakeup_flusher_threads(enum wb_reason reason);
188void wakeup_flusher_threads_bdi(struct backing_dev_info *bdi,
189 enum wb_reason reason);
190void inode_wait_for_writeback(struct inode *inode);
191void inode_io_list_del(struct inode *inode);
192
193/* writeback.h requires fs.h; it, too, is not included from here. */
194static inline void wait_on_inode(struct inode *inode)
195{
196 might_sleep();
197 wait_on_bit(&inode->i_state, __I_NEW, TASK_UNINTERRUPTIBLE);
198}
199
200#ifdef CONFIG_CGROUP_WRITEBACK
201
202#include <linux/cgroup.h>
203#include <linux/bio.h>
204
205void __inode_attach_wb(struct inode *inode, struct folio *folio);
206void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
207 struct inode *inode)
208 __releases(&inode->i_lock);
209void wbc_detach_inode(struct writeback_control *wbc);
210void wbc_account_cgroup_owner(struct writeback_control *wbc, struct page *page,
211 size_t bytes);
212int cgroup_writeback_by_id(u64 bdi_id, int memcg_id,
213 enum wb_reason reason, struct wb_completion *done);
214void cgroup_writeback_umount(void);
215bool cleanup_offline_cgwb(struct bdi_writeback *wb);
216
217/**
218 * inode_attach_wb - associate an inode with its wb
219 * @inode: inode of interest
220 * @folio: folio being dirtied (may be NULL)
221 *
222 * If @inode doesn't have its wb, associate it with the wb matching the
223 * memcg of @folio or, if @folio 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 folio *folio)
227{
228 if (!inode->i_wb)
229 __inode_attach_wb(inode, folio);
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) {
241 WARN_ON_ONCE(!(inode->i_state & I_CLEAR));
242 wb_put(inode->i_wb);
243 inode->i_wb = NULL;
244 }
245}
246
247/**
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. Must be called after the bio has been associated with
272 * a device.
273 */
274static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
275{
276 /*
277 * pageout() path doesn't attach @wbc to the inode being written
278 * out. This is intentional as we don't want the function to block
279 * behind a slow cgroup. Ultimately, we want pageout() to kick off
280 * regular writeback instead of writing things out itself.
281 */
282 if (wbc->wb)
283 bio_associate_blkg_from_css(bio, wbc->wb->blkcg_css);
284}
285
286#else /* CONFIG_CGROUP_WRITEBACK */
287
288static inline void inode_attach_wb(struct inode *inode, struct folio *folio)
289{
290}
291
292static inline void inode_detach_wb(struct inode *inode)
293{
294}
295
296static inline void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
297 struct inode *inode)
298 __releases(&inode->i_lock)
299{
300 spin_unlock(&inode->i_lock);
301}
302
303static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
304 struct inode *inode)
305{
306}
307
308static inline void wbc_detach_inode(struct writeback_control *wbc)
309{
310}
311
312static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
313{
314}
315
316static inline void wbc_account_cgroup_owner(struct writeback_control *wbc,
317 struct page *page, size_t bytes)
318{
319}
320
321static inline void cgroup_writeback_umount(void)
322{
323}
324
325#endif /* CONFIG_CGROUP_WRITEBACK */
326
327/*
328 * mm/page-writeback.c
329 */
330void laptop_io_completion(struct backing_dev_info *info);
331void laptop_sync_completion(void);
332void laptop_mode_timer_fn(struct timer_list *t);
333bool node_dirty_ok(struct pglist_data *pgdat);
334int wb_domain_init(struct wb_domain *dom, gfp_t gfp);
335#ifdef CONFIG_CGROUP_WRITEBACK
336void wb_domain_exit(struct wb_domain *dom);
337#endif
338
339extern struct wb_domain global_wb_domain;
340
341/* These are exported to sysctl. */
342extern unsigned int dirty_writeback_interval;
343extern unsigned int dirty_expire_interval;
344extern unsigned int dirtytime_expire_interval;
345extern int laptop_mode;
346
347int dirtytime_interval_handler(struct ctl_table *table, int write,
348 void *buffer, size_t *lenp, loff_t *ppos);
349
350void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty);
351unsigned long wb_calc_thresh(struct bdi_writeback *wb, unsigned long thresh);
352
353void wb_update_bandwidth(struct bdi_writeback *wb);
354
355/* Invoke balance dirty pages in async mode. */
356#define BDP_ASYNC 0x0001
357
358void balance_dirty_pages_ratelimited(struct address_space *mapping);
359int balance_dirty_pages_ratelimited_flags(struct address_space *mapping,
360 unsigned int flags);
361
362bool wb_over_bg_thresh(struct bdi_writeback *wb);
363
364typedef int (*writepage_t)(struct folio *folio, struct writeback_control *wbc,
365 void *data);
366
367void tag_pages_for_writeback(struct address_space *mapping,
368 pgoff_t start, pgoff_t end);
369int write_cache_pages(struct address_space *mapping,
370 struct writeback_control *wbc, writepage_t writepage,
371 void *data);
372int do_writepages(struct address_space *mapping, struct writeback_control *wbc);
373void writeback_set_ratelimit(void);
374void tag_pages_for_writeback(struct address_space *mapping,
375 pgoff_t start, pgoff_t end);
376
377bool filemap_dirty_folio(struct address_space *mapping, struct folio *folio);
378void folio_account_redirty(struct folio *folio);
379static inline void account_page_redirty(struct page *page)
380{
381 folio_account_redirty(page_folio(page));
382}
383bool folio_redirty_for_writepage(struct writeback_control *, struct folio *);
384bool redirty_page_for_writepage(struct writeback_control *, struct page *);
385
386void sb_mark_inode_writeback(struct inode *inode);
387void sb_clear_inode_writeback(struct inode *inode);
388
389#endif /* WRITEBACK_H */