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 #include <linux/blk_types.h>
  13
  14 struct bio;
  15
  16 DECLARE_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
  34 struct backing_dev_info;
  35
  36 /*
  37  * fs/fs-writeback.c
  38  */
  39 enum 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  */
  47 enum wb_reason {
  48         WB_REASON_BACKGROUND,
  49         WB_REASON_VMSCAN,
  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  */
  71 struct 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
 106 static 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  */
 123 struct 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  */
 173 static 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  */
 184 struct bdi_writeback;
 185 void writeback_inodes_sb(struct super_block *, enum wb_reason reason);
 186 void writeback_inodes_sb_nr(struct super_block *, unsigned long nr,
 187                                                         enum wb_reason reason);
 188 bool try_to_writeback_inodes_sb(struct super_block *, enum wb_reason reason);
 189 bool try_to_writeback_inodes_sb_nr(struct super_block *, unsigned long nr,
 190                                    enum wb_reason reason);
 191 void sync_inodes_sb(struct super_block *);
 192 void wakeup_flusher_threads(long nr_pages, enum wb_reason reason);
 193 void inode_wait_for_writeback(struct inode *inode);
 194
 195 /* writeback.h requires fs.h; it, too, is not included from here. */
 196 static 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
 207 void __inode_attach_wb(struct inode *inode, struct page *page);
 208 void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
 209                                  struct inode *inode)
 210         __releases(&inode->i_lock);
 211 void wbc_detach_inode(struct writeback_control *wbc);
 212 void wbc_account_io(struct writeback_control *wbc, struct page *page,
 213                     size_t bytes);
 214 void 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  */
 225 static 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  */
 237 static inline void inode_detach_wb(struct inode *inode)
 238 {
 239         if (inode->i_wb) {
 240                 WARN_ON_ONCE(!(inode->i_state & I_CLEAR));
 241                 wb_put(inode->i_wb);
 242                 inode->i_wb = NULL;
 243         }
 244 }
 245
 246 /**
 247  * wbc_attach_fdatawrite_inode - associate wbc and inode for fdatawrite
 248  * @wbc: writeback_control of interest
 249  * @inode: target inode
 250  *
 251  * This function is to be used by __filemap_fdatawrite_range(), which is an
 252  * alternative entry point into writeback code, and first ensures @inode is
 253  * associated with a bdi_writeback and attaches it to @wbc.
 254  */
 255 static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
 256                                                struct inode *inode)
 257 {
 258         spin_lock(&inode->i_lock);
 259         inode_attach_wb(inode, NULL);
 260         wbc_attach_and_unlock_inode(wbc, inode);
 261 }
 262
 263 /**
 264  * wbc_init_bio - writeback specific initializtion of bio
 265  * @wbc: writeback_control for the writeback in progress
 266  * @bio: bio to be initialized
 267  *
 268  * @bio is a part of the writeback in progress controlled by @wbc.  Perform
 269  * writeback specific initialization.  This is used to apply the cgroup
 270  * writeback context.
 271  */
 272 static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
 273 {
 274         /*
 275          * pageout() path doesn't attach @wbc to the inode being written
 276          * out.  This is intentional as we don't want the function to block
 277          * behind a slow cgroup.  Ultimately, we want pageout() to kick off
 278          * regular writeback instead of writing things out itself.
 279          */
 280         if (wbc->wb)
 281                 bio_associate_blkcg(bio, wbc->wb->blkcg_css);
 282 }
 283
 284 #else   /* CONFIG_CGROUP_WRITEBACK */
 285
 286 static inline void inode_attach_wb(struct inode *inode, struct page *page)
 287 {
 288 }
 289
 290 static inline void inode_detach_wb(struct inode *inode)
 291 {
 292 }
 293
 294 static inline void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
 295                                                struct inode *inode)
 296         __releases(&inode->i_lock)
 297 {
 298         spin_unlock(&inode->i_lock);
 299 }
 300
 301 static inline void wbc_attach_fdatawrite_inode(struct writeback_control *wbc,
 302                                                struct inode *inode)
 303 {
 304 }
 305
 306 static inline void wbc_detach_inode(struct writeback_control *wbc)
 307 {
 308 }
 309
 310 static inline void wbc_init_bio(struct writeback_control *wbc, struct bio *bio)
 311 {
 312 }
 313
 314 static inline void wbc_account_io(struct writeback_control *wbc,
 315                                   struct page *page, size_t bytes)
 316 {
 317 }
 318
 319 static inline void cgroup_writeback_umount(void)
 320 {
 321 }
 322
 323 #endif  /* CONFIG_CGROUP_WRITEBACK */
 324
 325 /*
 326  * mm/page-writeback.c
 327  */
 328 #ifdef CONFIG_BLOCK
 329 void laptop_io_completion(struct backing_dev_info *info);
 330 void laptop_sync_completion(void);
 331 void laptop_mode_sync(struct work_struct *work);
 332 void laptop_mode_timer_fn(unsigned long data);
 333 #else
 334 static inline void laptop_sync_completion(void) { }
 335 #endif
 336 bool node_dirty_ok(struct pglist_data *pgdat);
 337 int wb_domain_init(struct wb_domain *dom, gfp_t gfp);
 338 #ifdef CONFIG_CGROUP_WRITEBACK
 339 void wb_domain_exit(struct wb_domain *dom);
 340 #endif
 341
 342 extern struct wb_domain global_wb_domain;
 343
 344 /* These are exported to sysctl. */
 345 extern int dirty_background_ratio;
 346 extern unsigned long dirty_background_bytes;
 347 extern int vm_dirty_ratio;
 348 extern unsigned long vm_dirty_bytes;
 349 extern unsigned int dirty_writeback_interval;
 350 extern unsigned int dirty_expire_interval;
 351 extern unsigned int dirtytime_expire_interval;
 352 extern int vm_highmem_is_dirtyable;
 353 extern int block_dump;
 354 extern int laptop_mode;
 355
 356 extern int dirty_background_ratio_handler(struct ctl_table *table, int write,
 357                 void __user *buffer, size_t *lenp,
 358                 loff_t *ppos);
 359 extern int dirty_background_bytes_handler(struct ctl_table *table, int write,
 360                 void __user *buffer, size_t *lenp,
 361                 loff_t *ppos);
 362 extern int dirty_ratio_handler(struct ctl_table *table, int write,
 363                 void __user *buffer, size_t *lenp,
 364                 loff_t *ppos);
 365 extern int dirty_bytes_handler(struct ctl_table *table, int write,
 366                 void __user *buffer, size_t *lenp,
 367                 loff_t *ppos);
 368 int dirtytime_interval_handler(struct ctl_table *table, int write,
 369                                void __user *buffer, size_t *lenp, loff_t *ppos);
 370
 371 struct ctl_table;
 372 int dirty_writeback_centisecs_handler(struct ctl_table *, int,
 373                                       void __user *, size_t *, loff_t *);
 374
 375 void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty);
 376 unsigned long wb_calc_thresh(struct bdi_writeback *wb, unsigned long thresh);
 377
 378 void wb_update_bandwidth(struct bdi_writeback *wb, unsigned long start_time);
 379 void balance_dirty_pages_ratelimited(struct address_space *mapping);
 380 bool wb_over_bg_thresh(struct bdi_writeback *wb);
 381
 382 typedef int (*writepage_t)(struct page *page, struct writeback_control *wbc,
 383                                 void *data);
 384
 385 int generic_writepages(struct address_space *mapping,
 386                        struct writeback_control *wbc);
 387 void tag_pages_for_writeback(struct address_space *mapping,
 388                              pgoff_t start, pgoff_t end);
 389 int write_cache_pages(struct address_space *mapping,
 390                       struct writeback_control *wbc, writepage_t writepage,
 391                       void *data);
 392 int do_writepages(struct address_space *mapping, struct writeback_control *wbc);
 393 void writeback_set_ratelimit(void);
 394 void tag_pages_for_writeback(struct address_space *mapping,
 395                              pgoff_t start, pgoff_t end);
 396
 397 void account_page_redirty(struct page *page);
 398
 399 void sb_mark_inode_writeback(struct inode *inode);
 400 void sb_clear_inode_writeback(struct inode *inode);
 401
 402 #endif          /* WRITEBACK_H */