projects/linux·blob v4.19-rc6 · refs

381 lines · 12.1 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 1/4 region under the global dirty thresh is for smooth dirty throttling:
21 *
22 * (thresh - thresh/DIRTY_FULL_SCOPE, thresh)
23 *
24 * 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 */
32#define DIRTY_SCOPE 8
33#define DIRTY_FULL_SCOPE (DIRTY_SCOPE / 2)
34
35struct backing_dev_info;
36
37/*
38 * 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 */
43};
44
45/*
46 * A control structure which tells the writeback code what to do. These are
47 * always on the stack, and hence need no locking. They are always initialised
48 * in a manner such that unspecified fields are set to zero.
49 */
50struct writeback_control {
51 long nr_to_write; /* Write this many pages, and decrement
52 this for each page written */
53 long pages_skipped; /* Pages which were not written */
54
55 /*
56 * For a_ops->writepages(): if start or end are non-zero then this is
57 * a hint that the filesystem need only write out the pages inside that
58 * byterange. The byte at `end' is included in the writeout request.
59 */
60 loff_t range_start;
61 loff_t range_end;
62
63 enum writeback_sync_modes sync_mode;
64
65 unsigned for_kupdate:1; /* A kupdate writeback */
66 unsigned for_background:1; /* A background writeback */
67 unsigned tagged_writepages:1; /* tag-and-write to avoid livelock */
68 unsigned for_reclaim:1; /* Invoked from the page allocator */
69 unsigned range_cyclic:1; /* range_start is cyclic */
70 unsigned for_sync:1; /* sync(2) WB_SYNC_ALL writeback */
71#ifdef CONFIG_CGROUP_WRITEBACK
72 struct bdi_writeback *wb; /* wb this writeback is issued under */
73 struct inode *inode; /* inode being written out */
74
75 /* foreign inode detection, see wbc_detach_inode() */
76 int wb_id; /* current wb id */
77 int wb_lcand_id; /* last foreign candidate wb id */
78 int wb_tcand_id; /* this foreign candidate wb id */
79 size_t wb_bytes; /* bytes written by current wb */
80 size_t wb_lcand_bytes; /* bytes written by last candidate */
81 size_t wb_tcand_bytes; /* bytes written by this candidate */
82#endif
83};
84
85static inline int wbc_to_write_flags(struct writeback_control *wbc)
86{
87 if (wbc->sync_mode == WB_SYNC_ALL)
88 return REQ_SYNC;
89 else if (wbc->for_kupdate || wbc->for_background)
90 return REQ_BACKGROUND;
91
92 return 0;
93}
94
95/*
96 * A wb_domain represents a domain that wb's (bdi_writeback's) belong to
97 * and are measured against each other in. There always is one global
98 * domain, global_wb_domain, that every wb in the system is a member of.
99 * This allows measuring the relative bandwidth of each wb to distribute
100 * dirtyable memory accordingly.
101 */
102struct wb_domain {
103 spinlock_t lock;
104
105 /*
106 * Scale the writeback cache size proportional to the relative
107 * writeout speed.
108 *
109 * We do this by keeping a floating proportion between BDIs, based
110 * on page writeback completions [end_page_writeback()]. Those
111 * devices that write out pages fastest will get the larger share,
112 * while the slower will get a smaller share.
113 *
114 * We use page writeout completions because we are interested in
115 * getting rid of dirty pages. Having them written out is the
116 * primary goal.
117 *
118 * We introduce a concept of time, a period over which we measure
119 * these events, because demand can/will vary over time. The length
120 * of this period itself is measured in page writeback completions.
121 */
122 struct fprop_global completions;
123 struct timer_list period_timer; /* timer for aging of completions */
124 unsigned long period_time;
125
126 /*
127 * The dirtyable memory and dirty threshold could be suddenly
128 * knocked down by a large amount (eg. on the startup of KVM in a
129 * swapless system). This may throw the system into deep dirty
130 * exceeded state and throttle heavy/light dirtiers alike. To
131 * retain good responsiveness, maintain global_dirty_limit for
132 * tracking slowly down to the knocked down dirty threshold.
133 *
134 * Both fields are protected by ->lock.
135 */
136 unsigned long dirty_limit_tstamp;
137 unsigned long dirty_limit;
138};
139
140/**
141 * wb_domain_size_changed - memory available to a wb_domain has changed
142 * @dom: wb_domain of interest
143 *
144 * This function should be called when the amount of memory available to
145 * @dom has changed. It resets @dom's dirty limit parameters to prevent
146 * the past values which don't match the current configuration from skewing
147 * dirty throttling. Without this, when memory size of a wb_domain is
148 * greatly reduced, the dirty throttling logic may allow too many pages to
149 * be dirtied leading to consecutive unnecessary OOMs and may get stuck in
150 * that situation.
151 */
152static inline void wb_domain_size_changed(struct wb_domain *dom)
153{
154 spin_lock(&dom->lock);
155 dom->dirty_limit_tstamp = jiffies;
156 dom->dirty_limit = 0;
157 spin_unlock(&dom->lock);
158}
159
160/*
161 * fs/fs-writeback.c
162 */
163struct bdi_writeback;
164void writeback_inodes_sb(struct super_block *, enum wb_reason reason);
165void writeback_inodes_sb_nr(struct super_block *, unsigned long nr,
166 enum wb_reason reason);
167void try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason);
168void sync_inodes_sb(struct super_block *);
169void wakeup_flusher_threads(enum wb_reason reason);
170void wakeup_flusher_threads_bdi(struct backing_dev_info *bdi,
171 enum wb_reason reason);
172void inode_wait_for_writeback(struct inode *inode);
173
174/* writeback.h requires fs.h; it, too, is not included from here. */
175static inline void wait_on_inode(struct inode *inode)
176{
177 might_sleep();
178 wait_on_bit(&inode->i_state, __I_NEW, TASK_UNINTERRUPTIBLE);
179}
180
181#ifdef CONFIG_CGROUP_WRITEBACK
182
183#include <linux/cgroup.h>
184#include <linux/bio.h>
185
186void __inode_attach_wb(struct inode *inode, struct page *page);
187void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
188 struct inode *inode)
189 __releases(&inode->i_lock);
190void wbc_detach_inode(struct writeback_control *wbc);
191void wbc_account_io(struct writeback_control *wbc, struct page *page,
192 size_t bytes);
193void cgroup_writeback_umount(void);
194
195/**
196 * inode_attach_wb - associate an inode with its wb
197 * @inode: inode of interest
198 * @page: page being dirtied (may be NULL)
199 *
200 * If @inode doesn't have its wb, associate it with the wb matching the
201 * memcg of @page or, if @page is NULL, %current. May be called w/ or w/o
202 * @inode->i_lock.
203 */
204static inline void inode_attach_wb(struct inode *inode, struct page *page)
205{
206 if (!inode->i_wb)
207 __inode_attach_wb(inode, page);
208}
209
210/**
211 * inode_detach_wb - disassociate an inode from its wb
212 * @inode: inode of interest
213 *
214 * @inode is being freed. Detach from its wb.
215 */
216static inline void inode_detach_wb(struct inode *inode)
217{
218 if (inode->i_wb) {
219 WARN_ON_ONCE(!(inode->i_state & I_CLEAR));
220 wb_put(inode->i_wb);
221 inode->i_wb = NULL;
222 }
223}
224
225/**
226 * wbc_attach_fdatawrite_inode - associate wbc and inode for fdatawrite
227 * @wbc: writeback_control of interest
228 * @inode: target inode
229 *
230 * This function is to be used by __filemap_fdatawrite_range(), which is an
231 * alternative entry point into writeback code, and first ensures @inode is
232 * associated with a bdi_writeback and attaches it to @wbc.
233 */
234static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
235 struct inode *inode)
236{
237 spin_lock(&inode->i_lock);
238 inode_attach_wb(inode, NULL);
239 wbc_attach_and_unlock_inode(wbc, inode);
240}
241
242/**
243 * wbc_init_bio - writeback specific initializtion of bio
244 * @wbc: writeback_control for the writeback in progress
245 * @bio: bio to be initialized
246 *
247 * @bio is a part of the writeback in progress controlled by @wbc. Perform
248 * writeback specific initialization. This is used to apply the cgroup
249 * writeback context.
250 */
251static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
252{
253 /*
254 * pageout() path doesn't attach @wbc to the inode being written
255 * out. This is intentional as we don't want the function to block
256 * behind a slow cgroup. Ultimately, we want pageout() to kick off
257 * regular writeback instead of writing things out itself.
258 */
259 if (wbc->wb)
260 bio_associate_blkcg(bio, wbc->wb->blkcg_css);
261}
262
263#else /* CONFIG_CGROUP_WRITEBACK */
264
265static inline void inode_attach_wb(struct inode *inode, struct page *page)
266{
267}
268
269static inline void inode_detach_wb(struct inode *inode)
270{
271}
272
273static inline void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
274 struct inode *inode)
275 __releases(&inode->i_lock)
276{
277 spin_unlock(&inode->i_lock);
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_io(struct writeback_control *wbc,
294 struct page *page, size_t bytes)
295{
296}
297
298static inline void cgroup_writeback_umount(void)
299{
300}
301
302#endif /* CONFIG_CGROUP_WRITEBACK */
303
304/*
305 * mm/page-writeback.c
306 */
307#ifdef CONFIG_BLOCK
308void laptop_io_completion(struct backing_dev_info *info);
309void laptop_sync_completion(void);
310void laptop_mode_sync(struct work_struct *work);
311void laptop_mode_timer_fn(struct timer_list *t);
312#else
313static inline void laptop_sync_completion(void) { }
314#endif
315bool node_dirty_ok(struct pglist_data *pgdat);
316int wb_domain_init(struct wb_domain *dom, gfp_t gfp);
317#ifdef CONFIG_CGROUP_WRITEBACK
318void wb_domain_exit(struct wb_domain *dom);
319#endif
320
321extern struct wb_domain global_wb_domain;
322
323/* These are exported to sysctl. */
324extern int dirty_background_ratio;
325extern unsigned long dirty_background_bytes;
326extern int vm_dirty_ratio;
327extern unsigned long vm_dirty_bytes;
328extern unsigned int dirty_writeback_interval;
329extern unsigned int dirty_expire_interval;
330extern unsigned int dirtytime_expire_interval;
331extern int vm_highmem_is_dirtyable;
332extern int block_dump;
333extern int laptop_mode;
334
335extern int dirty_background_ratio_handler(struct ctl_table *table, int write,
336 void __user *buffer, size_t *lenp,
337 loff_t *ppos);
338extern int dirty_background_bytes_handler(struct ctl_table *table, int write,
339 void __user *buffer, size_t *lenp,
340 loff_t *ppos);
341extern int dirty_ratio_handler(struct ctl_table *table, int write,
342 void __user *buffer, size_t *lenp,
343 loff_t *ppos);
344extern int dirty_bytes_handler(struct ctl_table *table, int write,
345 void __user *buffer, size_t *lenp,
346 loff_t *ppos);
347int dirtytime_interval_handler(struct ctl_table *table, int write,
348 void __user *buffer, size_t *lenp, loff_t *ppos);
349
350struct ctl_table;
351int dirty_writeback_centisecs_handler(struct ctl_table *, int,
352 void __user *, size_t *, loff_t *);
353
354void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty);
355unsigned long wb_calc_thresh(struct bdi_writeback *wb, unsigned long thresh);
356
357void wb_update_bandwidth(struct bdi_writeback *wb, unsigned long start_time);
358void balance_dirty_pages_ratelimited(struct address_space *mapping);
359bool wb_over_bg_thresh(struct bdi_writeback *wb);
360
361typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
362 void *data);
363
364int generic_writepages(struct address_space *mapping,
365 struct writeback_control *wbc);
366void tag_pages_for_writeback(struct address_space *mapping,
367 pgoff_t start, pgoff_t end);
368int write_cache_pages(struct address_space *mapping,
369 struct writeback_control *wbc, writepage_t writepage,
370 void *data);
371int do_writepages(struct address_space *mapping, struct writeback_control *wbc);
372void writeback_set_ratelimit(void);
373void tag_pages_for_writeback(struct address_space *mapping,
374 pgoff_t start, pgoff_t end);
375
376void account_page_redirty(struct page *page);
377
378void sb_mark_inode_writeback(struct inode *inode);
379void sb_clear_inode_writeback(struct inode *inode);
380
381#endif /* WRITEBACK_H */