include/linux/writeback.h

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