projects/linux·blob v4.10-rc7 · refs

401 lines · 12.6 KiB

1/*
2 * include/linux/writeback.h
3 */
4#ifndef WRITEBACK_H
5#define WRITEBACK_H
6
7#include <linux/sched.h>
8#include <linux/workqueue.h>
9#include <linux/fs.h>
10#include <linux/flex_proportions.h>
11#include <linux/backing-dev-defs.h>
12#include <linux/blk_types.h>
13
14struct bio;
15
16DECLARE_PER_CPU(int, dirty_throttle_leaks);
17
18/*
19 * The 1/4 region under the global dirty thresh is for smooth dirty throttling:
20 *
21 * (thresh - thresh/DIRTY_FULL_SCOPE, thresh)
22 *
23 * Further beyond, all dirtier tasks will enter a loop waiting (possibly long
24 * time) for the dirty pages to drop, unless written enough pages.
25 *
26 * The global dirty threshold is normally equal to the global dirty limit,
27 * except when the system suddenly allocates a lot of anonymous memory and
28 * knocks down the global dirty threshold quickly, in which case the global
29 * dirty limit will follow down slowly to prevent livelocking all dirtier tasks.
30 */
31#define DIRTY_SCOPE 8
32#define DIRTY_FULL_SCOPE (DIRTY_SCOPE / 2)
33
34struct backing_dev_info;
35
36/*
37 * fs/fs-writeback.c
38 */
39enum writeback_sync_modes {
40 WB_SYNC_NONE, /* Don't wait on anything */
41 WB_SYNC_ALL, /* Wait on every mapping */
42};
43
44/*
45 * why some writeback work was initiated
46 */
47enum wb_reason {
48 WB_REASON_BACKGROUND,
49 WB_REASON_TRY_TO_FREE_PAGES,
50 WB_REASON_SYNC,
51 WB_REASON_PERIODIC,
52 WB_REASON_LAPTOP_TIMER,
53 WB_REASON_FREE_MORE_MEM,
54 WB_REASON_FS_FREE_SPACE,
55 /*
56 * There is no bdi forker thread any more and works are done
57 * by emergency worker, however, this is TPs userland visible
58 * and we'll be exposing exactly the same information,
59 * so it has a mismatch name.
60 */
61 WB_REASON_FORKER_THREAD,
62
63 WB_REASON_MAX,
64};
65
66/*
67 * A control structure which tells the writeback code what to do. These are
68 * always on the stack, and hence need no locking. They are always initialised
69 * in a manner such that unspecified fields are set to zero.
70 */
71struct writeback_control {
72 long nr_to_write; /* Write this many pages, and decrement
73 this for each page written */
74 long pages_skipped; /* Pages which were not written */
75
76 /*
77 * For a_ops->writepages(): if start or end are non-zero then this is
78 * a hint that the filesystem need only write out the pages inside that
79 * byterange. The byte at `end' is included in the writeout request.
80 */
81 loff_t range_start;
82 loff_t range_end;
83
84 enum writeback_sync_modes sync_mode;
85
86 unsigned for_kupdate:1; /* A kupdate writeback */
87 unsigned for_background:1; /* A background writeback */
88 unsigned tagged_writepages:1; /* tag-and-write to avoid livelock */
89 unsigned for_reclaim:1; /* Invoked from the page allocator */
90 unsigned range_cyclic:1; /* range_start is cyclic */
91 unsigned for_sync:1; /* sync(2) WB_SYNC_ALL writeback */
92#ifdef CONFIG_CGROUP_WRITEBACK
93 struct bdi_writeback *wb; /* wb this writeback is issued under */
94 struct inode *inode; /* inode being written out */
95
96 /* foreign inode detection, see wbc_detach_inode() */
97 int wb_id; /* current wb id */
98 int wb_lcand_id; /* last foreign candidate wb id */
99 int wb_tcand_id; /* this foreign candidate wb id */
100 size_t wb_bytes; /* bytes written by current wb */
101 size_t wb_lcand_bytes; /* bytes written by last candidate */
102 size_t wb_tcand_bytes; /* bytes written by this candidate */
103#endif
104};
105
106static inline int wbc_to_write_flags(struct writeback_control *wbc)
107{
108 if (wbc->sync_mode == WB_SYNC_ALL)
109 return REQ_SYNC;
110 else if (wbc->for_kupdate || wbc->for_background)
111 return REQ_BACKGROUND;
112
113 return 0;
114}
115
116/*
117 * A wb_domain represents a domain that wb's (bdi_writeback's) belong to
118 * and are measured against each other in. There always is one global
119 * domain, global_wb_domain, that every wb in the system is a member of.
120 * This allows measuring the relative bandwidth of each wb to distribute
121 * dirtyable memory accordingly.
122 */
123struct wb_domain {
124 spinlock_t lock;
125
126 /*
127 * Scale the writeback cache size proportional to the relative
128 * writeout speed.
129 *
130 * We do this by keeping a floating proportion between BDIs, based
131 * on page writeback completions [end_page_writeback()]. Those
132 * devices that write out pages fastest will get the larger share,
133 * while the slower will get a smaller share.
134 *
135 * We use page writeout completions because we are interested in
136 * getting rid of dirty pages. Having them written out is the
137 * primary goal.
138 *
139 * We introduce a concept of time, a period over which we measure
140 * these events, because demand can/will vary over time. The length
141 * of this period itself is measured in page writeback completions.
142 */
143 struct fprop_global completions;
144 struct timer_list period_timer; /* timer for aging of completions */
145 unsigned long period_time;
146
147 /*
148 * The dirtyable memory and dirty threshold could be suddenly
149 * knocked down by a large amount (eg. on the startup of KVM in a
150 * swapless system). This may throw the system into deep dirty
151 * exceeded state and throttle heavy/light dirtiers alike. To
152 * retain good responsiveness, maintain global_dirty_limit for
153 * tracking slowly down to the knocked down dirty threshold.
154 *
155 * Both fields are protected by ->lock.
156 */
157 unsigned long dirty_limit_tstamp;
158 unsigned long dirty_limit;
159};
160
161/**
162 * wb_domain_size_changed - memory available to a wb_domain has changed
163 * @dom: wb_domain of interest
164 *
165 * This function should be called when the amount of memory available to
166 * @dom has changed. It resets @dom's dirty limit parameters to prevent
167 * the past values which don't match the current configuration from skewing
168 * dirty throttling. Without this, when memory size of a wb_domain is
169 * greatly reduced, the dirty throttling logic may allow too many pages to
170 * be dirtied leading to consecutive unnecessary OOMs and may get stuck in
171 * that situation.
172 */
173static inline void wb_domain_size_changed(struct wb_domain *dom)
174{
175 spin_lock(&dom->lock);
176 dom->dirty_limit_tstamp = jiffies;
177 dom->dirty_limit = 0;
178 spin_unlock(&dom->lock);
179}
180
181/*
182 * fs/fs-writeback.c
183 */
184struct bdi_writeback;
185void writeback_inodes_sb(struct super_block *, enum wb_reason reason);
186void writeback_inodes_sb_nr(struct super_block *, unsigned long nr,
187 enum wb_reason reason);
188bool try_to_writeback_inodes_sb(struct super_block *, enum wb_reason reason);
189bool try_to_writeback_inodes_sb_nr(struct super_block *, unsigned long nr,
190 enum wb_reason reason);
191void sync_inodes_sb(struct super_block *);
192void wakeup_flusher_threads(long nr_pages, enum wb_reason reason);
193void inode_wait_for_writeback(struct inode *inode);
194
195/* writeback.h requires fs.h; it, too, is not included from here. */
196static inline void wait_on_inode(struct inode *inode)
197{
198 might_sleep();
199 wait_on_bit(&inode->i_state, __I_NEW, TASK_UNINTERRUPTIBLE);
200}
201
202#ifdef CONFIG_CGROUP_WRITEBACK
203
204#include <linux/cgroup.h>
205#include <linux/bio.h>
206
207void __inode_attach_wb(struct inode *inode, struct page *page);
208void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
209 struct inode *inode)
210 __releases(&inode->i_lock);
211void wbc_detach_inode(struct writeback_control *wbc);
212void wbc_account_io(struct writeback_control *wbc, struct page *page,
213 size_t bytes);
214void cgroup_writeback_umount(void);
215
216/**
217 * inode_attach_wb - associate an inode with its wb
218 * @inode: inode of interest
219 * @page: page being dirtied (may be NULL)
220 *
221 * If @inode doesn't have its wb, associate it with the wb matching the
222 * memcg of @page or, if @page is NULL, %current. May be called w/ or w/o
223 * @inode->i_lock.
224 */
225static inline void inode_attach_wb(struct inode *inode, struct page *page)
226{
227 if (!inode->i_wb)
228 __inode_attach_wb(inode, page);
229}
230
231/**
232 * inode_detach_wb - disassociate an inode from its wb
233 * @inode: inode of interest
234 *
235 * @inode is being freed. Detach from its wb.
236 */
237static inline void inode_detach_wb(struct inode *inode)
238{
239 if (inode->i_wb) {
240 wb_put(inode->i_wb);
241 inode->i_wb = NULL;
242 }
243}
244
245/**
246 * wbc_attach_fdatawrite_inode - associate wbc and inode for fdatawrite
247 * @wbc: writeback_control of interest
248 * @inode: target inode
249 *
250 * This function is to be used by __filemap_fdatawrite_range(), which is an
251 * alternative entry point into writeback code, and first ensures @inode is
252 * associated with a bdi_writeback and attaches it to @wbc.
253 */
254static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
255 struct inode *inode)
256{
257 spin_lock(&inode->i_lock);
258 inode_attach_wb(inode, NULL);
259 wbc_attach_and_unlock_inode(wbc, inode);
260}
261
262/**
263 * wbc_init_bio - writeback specific initializtion of bio
264 * @wbc: writeback_control for the writeback in progress
265 * @bio: bio to be initialized
266 *
267 * @bio is a part of the writeback in progress controlled by @wbc. Perform
268 * writeback specific initialization. This is used to apply the cgroup
269 * writeback context.
270 */
271static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
272{
273 /*
274 * pageout() path doesn't attach @wbc to the inode being written
275 * out. This is intentional as we don't want the function to block
276 * behind a slow cgroup. Ultimately, we want pageout() to kick off
277 * regular writeback instead of writing things out itself.
278 */
279 if (wbc->wb)
280 bio_associate_blkcg(bio, wbc->wb->blkcg_css);
281}
282
283#else /* CONFIG_CGROUP_WRITEBACK */
284
285static inline void inode_attach_wb(struct inode *inode, struct page *page)
286{
287}
288
289static inline void inode_detach_wb(struct inode *inode)
290{
291}
292
293static inline void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
294 struct inode *inode)
295 __releases(&inode->i_lock)
296{
297 spin_unlock(&inode->i_lock);
298}
299
300static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
301 struct inode *inode)
302{
303}
304
305static inline void wbc_detach_inode(struct writeback_control *wbc)
306{
307}
308
309static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
310{
311}
312
313static inline void wbc_account_io(struct writeback_control *wbc,
314 struct page *page, size_t bytes)
315{
316}
317
318static inline void cgroup_writeback_umount(void)
319{
320}
321
322#endif /* CONFIG_CGROUP_WRITEBACK */
323
324/*
325 * mm/page-writeback.c
326 */
327#ifdef CONFIG_BLOCK
328void laptop_io_completion(struct backing_dev_info *info);
329void laptop_sync_completion(void);
330void laptop_mode_sync(struct work_struct *work);
331void laptop_mode_timer_fn(unsigned long data);
332#else
333static inline void laptop_sync_completion(void) { }
334#endif
335bool node_dirty_ok(struct pglist_data *pgdat);
336int wb_domain_init(struct wb_domain *dom, gfp_t gfp);
337#ifdef CONFIG_CGROUP_WRITEBACK
338void wb_domain_exit(struct wb_domain *dom);
339#endif
340
341extern struct wb_domain global_wb_domain;
342
343/* These are exported to sysctl. */
344extern int dirty_background_ratio;
345extern unsigned long dirty_background_bytes;
346extern int vm_dirty_ratio;
347extern unsigned long vm_dirty_bytes;
348extern unsigned int dirty_writeback_interval;
349extern unsigned int dirty_expire_interval;
350extern unsigned int dirtytime_expire_interval;
351extern int vm_highmem_is_dirtyable;
352extern int block_dump;
353extern int laptop_mode;
354
355extern int dirty_background_ratio_handler(struct ctl_table *table, int write,
356 void __user *buffer, size_t *lenp,
357 loff_t *ppos);
358extern int dirty_background_bytes_handler(struct ctl_table *table, int write,
359 void __user *buffer, size_t *lenp,
360 loff_t *ppos);
361extern int dirty_ratio_handler(struct ctl_table *table, int write,
362 void __user *buffer, size_t *lenp,
363 loff_t *ppos);
364extern int dirty_bytes_handler(struct ctl_table *table, int write,
365 void __user *buffer, size_t *lenp,
366 loff_t *ppos);
367int dirtytime_interval_handler(struct ctl_table *table, int write,
368 void __user *buffer, size_t *lenp, loff_t *ppos);
369
370struct ctl_table;
371int dirty_writeback_centisecs_handler(struct ctl_table *, int,
372 void __user *, size_t *, loff_t *);
373
374void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty);
375unsigned long wb_calc_thresh(struct bdi_writeback *wb, unsigned long thresh);
376
377void wb_update_bandwidth(struct bdi_writeback *wb, unsigned long start_time);
378void balance_dirty_pages_ratelimited(struct address_space *mapping);
379bool wb_over_bg_thresh(struct bdi_writeback *wb);
380
381typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
382 void *data);
383
384int generic_writepages(struct address_space *mapping,
385 struct writeback_control *wbc);
386void tag_pages_for_writeback(struct address_space *mapping,
387 pgoff_t start, pgoff_t end);
388int write_cache_pages(struct address_space *mapping,
389 struct writeback_control *wbc, writepage_t writepage,
390 void *data);
391int do_writepages(struct address_space *mapping, struct writeback_control *wbc);
392void writeback_set_ratelimit(void);
393void tag_pages_for_writeback(struct address_space *mapping,
394 pgoff_t start, pgoff_t end);
395
396void account_page_redirty(struct page *page);
397
398void sb_mark_inode_writeback(struct inode *inode);
399void sb_clear_inode_writeback(struct inode *inode);
400
401#endif /* WRITEBACK_H */