include/linux/backing-dev.h

   1 /*
   2  * include/linux/backing-dev.h
   3  *
   4  * low-level device information and state which is propagated up through
   5  * to high-level code.
   6  */
   7
   8 #ifndef _LINUX_BACKING_DEV_H
   9 #define _LINUX_BACKING_DEV_H
  10
  11 #include <linux/kernel.h>
  12 #include <linux/fs.h>
  13 #include <linux/sched.h>
  14 #include <linux/blkdev.h>
  15 #include <linux/writeback.h>
  16 #include <linux/blk-cgroup.h>
  17 #include <linux/backing-dev-defs.h>
  18
  19 int __must_check bdi_init(struct backing_dev_info *bdi);
  20 void bdi_destroy(struct backing_dev_info *bdi);
  21
  22 __printf(3, 4)
  23 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
  24                 const char *fmt, ...);
  25 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev);
  26 int __must_check bdi_setup_and_register(struct backing_dev_info *, char *);
  27 void wb_start_writeback(struct bdi_writeback *wb, long nr_pages,
  28                         bool range_cyclic, enum wb_reason reason);
  29 void wb_start_background_writeback(struct bdi_writeback *wb);
  30 void wb_workfn(struct work_struct *work);
  31 void wb_wakeup_delayed(struct bdi_writeback *wb);
  32
  33 extern spinlock_t bdi_lock;
  34 extern struct list_head bdi_list;
  35
  36 extern struct workqueue_struct *bdi_wq;
  37
  38 static inline bool wb_has_dirty_io(struct bdi_writeback *wb)
  39 {
  40         return test_bit(WB_has_dirty_io, &wb->state);
  41 }
  42
  43 static inline bool bdi_has_dirty_io(struct backing_dev_info *bdi)
  44 {
  45         /*
  46          * @bdi->tot_write_bandwidth is guaranteed to be > 0 if there are
  47          * any dirty wbs.  See wb_update_write_bandwidth().
  48          */
  49         return atomic_long_read(&bdi->tot_write_bandwidth);
  50 }
  51
  52 static inline void __add_wb_stat(struct bdi_writeback *wb,
  53                                  enum wb_stat_item item, s64 amount)
  54 {
  55         __percpu_counter_add(&wb->stat[item], amount, WB_STAT_BATCH);
  56 }
  57
  58 static inline void __inc_wb_stat(struct bdi_writeback *wb,
  59                                  enum wb_stat_item item)
  60 {
  61         __add_wb_stat(wb, item, 1);
  62 }
  63
  64 static inline void inc_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
  65 {
  66         unsigned long flags;
  67
  68         local_irq_save(flags);
  69         __inc_wb_stat(wb, item);
  70         local_irq_restore(flags);
  71 }
  72
  73 static inline void __dec_wb_stat(struct bdi_writeback *wb,
  74                                  enum wb_stat_item item)
  75 {
  76         __add_wb_stat(wb, item, -1);
  77 }
  78
  79 static inline void dec_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
  80 {
  81         unsigned long flags;
  82
  83         local_irq_save(flags);
  84         __dec_wb_stat(wb, item);
  85         local_irq_restore(flags);
  86 }
  87
  88 static inline s64 wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
  89 {
  90         return percpu_counter_read_positive(&wb->stat[item]);
  91 }
  92
  93 static inline s64 __wb_stat_sum(struct bdi_writeback *wb,
  94                                 enum wb_stat_item item)
  95 {
  96         return percpu_counter_sum_positive(&wb->stat[item]);
  97 }
  98
  99 static inline s64 wb_stat_sum(struct bdi_writeback *wb, enum wb_stat_item item)
 100 {
 101         s64 sum;
 102         unsigned long flags;
 103
 104         local_irq_save(flags);
 105         sum = __wb_stat_sum(wb, item);
 106         local_irq_restore(flags);
 107
 108         return sum;
 109 }
 110
 111 extern void wb_writeout_inc(struct bdi_writeback *wb);
 112
 113 /*
 114  * maximal error of a stat counter.
 115  */
 116 static inline unsigned long wb_stat_error(struct bdi_writeback *wb)
 117 {
 118 #ifdef CONFIG_SMP
 119         return nr_cpu_ids * WB_STAT_BATCH;
 120 #else
 121         return 1;
 122 #endif
 123 }
 124
 125 int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
 126 int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
 127
 128 /*
 129  * Flags in backing_dev_info::capability
 130  *
 131  * The first three flags control whether dirty pages will contribute to the
 132  * VM's accounting and whether writepages() should be called for dirty pages
 133  * (something that would not, for example, be appropriate for ramfs)
 134  *
 135  * WARNING: these flags are closely related and should not normally be
 136  * used separately.  The BDI_CAP_NO_ACCT_AND_WRITEBACK combines these
 137  * three flags into a single convenience macro.
 138  *
 139  * BDI_CAP_NO_ACCT_DIRTY:  Dirty pages shouldn't contribute to accounting
 140  * BDI_CAP_NO_WRITEBACK:   Don't write pages back
 141  * BDI_CAP_NO_ACCT_WB:     Don't automatically account writeback pages
 142  * BDI_CAP_STRICTLIMIT:    Keep number of dirty pages below bdi threshold.
 143  *
 144  * BDI_CAP_CGROUP_WRITEBACK: Supports cgroup-aware writeback.
 145  */
 146 #define BDI_CAP_NO_ACCT_DIRTY   0x00000001
 147 #define BDI_CAP_NO_WRITEBACK    0x00000002
 148 #define BDI_CAP_NO_ACCT_WB      0x00000004
 149 #define BDI_CAP_STABLE_WRITES   0x00000008
 150 #define BDI_CAP_STRICTLIMIT     0x00000010
 151 #define BDI_CAP_CGROUP_WRITEBACK 0x00000020
 152
 153 #define BDI_CAP_NO_ACCT_AND_WRITEBACK \
 154         (BDI_CAP_NO_WRITEBACK | BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_ACCT_WB)
 155
 156 extern struct backing_dev_info noop_backing_dev_info;
 157
 158 /**
 159  * writeback_in_progress - determine whether there is writeback in progress
 160  * @wb: bdi_writeback of interest
 161  *
 162  * Determine whether there is writeback waiting to be handled against a
 163  * bdi_writeback.
 164  */
 165 static inline bool writeback_in_progress(struct bdi_writeback *wb)
 166 {
 167         return test_bit(WB_writeback_running, &wb->state);
 168 }
 169
 170 static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
 171 {
 172         struct super_block *sb;
 173
 174         if (!inode)
 175                 return &noop_backing_dev_info;
 176
 177         sb = inode->i_sb;
 178 #ifdef CONFIG_BLOCK
 179         if (sb_is_blkdev_sb(sb))
 180                 return blk_get_backing_dev_info(I_BDEV(inode));
 181 #endif
 182         return sb->s_bdi;
 183 }
 184
 185 static inline int wb_congested(struct bdi_writeback *wb, int cong_bits)
 186 {
 187         struct backing_dev_info *bdi = wb->bdi;
 188
 189         if (bdi->congested_fn)
 190                 return bdi->congested_fn(bdi->congested_data, cong_bits);
 191         return wb->congested->state & cong_bits;
 192 }
 193
 194 long congestion_wait(int sync, long timeout);
 195 long wait_iff_congested(struct zone *zone, int sync, long timeout);
 196 int pdflush_proc_obsolete(struct ctl_table *table, int write,
 197                 void __user *buffer, size_t *lenp, loff_t *ppos);
 198
 199 static inline bool bdi_cap_stable_pages_required(struct backing_dev_info *bdi)
 200 {
 201         return bdi->capabilities & BDI_CAP_STABLE_WRITES;
 202 }
 203
 204 static inline bool bdi_cap_writeback_dirty(struct backing_dev_info *bdi)
 205 {
 206         return !(bdi->capabilities & BDI_CAP_NO_WRITEBACK);
 207 }
 208
 209 static inline bool bdi_cap_account_dirty(struct backing_dev_info *bdi)
 210 {
 211         return !(bdi->capabilities & BDI_CAP_NO_ACCT_DIRTY);
 212 }
 213
 214 static inline bool bdi_cap_account_writeback(struct backing_dev_info *bdi)
 215 {
 216         /* Paranoia: BDI_CAP_NO_WRITEBACK implies BDI_CAP_NO_ACCT_WB */
 217         return !(bdi->capabilities & (BDI_CAP_NO_ACCT_WB |
 218                                       BDI_CAP_NO_WRITEBACK));
 219 }
 220
 221 static inline bool mapping_cap_writeback_dirty(struct address_space *mapping)
 222 {
 223         return bdi_cap_writeback_dirty(inode_to_bdi(mapping->host));
 224 }
 225
 226 static inline bool mapping_cap_account_dirty(struct address_space *mapping)
 227 {
 228         return bdi_cap_account_dirty(inode_to_bdi(mapping->host));
 229 }
 230
 231 static inline int bdi_sched_wait(void *word)
 232 {
 233         schedule();
 234         return 0;
 235 }
 236
 237 #ifdef CONFIG_CGROUP_WRITEBACK
 238
 239 struct bdi_writeback_congested *
 240 wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp);
 241 void wb_congested_put(struct bdi_writeback_congested *congested);
 242 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
 243                                     struct cgroup_subsys_state *memcg_css,
 244                                     gfp_t gfp);
 245 void wb_memcg_offline(struct mem_cgroup *memcg);
 246 void wb_blkcg_offline(struct blkcg *blkcg);
 247 int inode_congested(struct inode *inode, int cong_bits);
 248
 249 /**
 250  * inode_cgwb_enabled - test whether cgroup writeback is enabled on an inode
 251  * @inode: inode of interest
 252  *
 253  * cgroup writeback requires support from both the bdi and filesystem.
 254  * Test whether @inode has both.
 255  */
 256 static inline bool inode_cgwb_enabled(struct inode *inode)
 257 {
 258         struct backing_dev_info *bdi = inode_to_bdi(inode);
 259
 260         return bdi_cap_account_dirty(bdi) &&
 261                 (bdi->capabilities & BDI_CAP_CGROUP_WRITEBACK) &&
 262                 (inode->i_sb->s_iflags & SB_I_CGROUPWB);
 263 }
 264
 265 /**
 266  * wb_find_current - find wb for %current on a bdi
 267  * @bdi: bdi of interest
 268  *
 269  * Find the wb of @bdi which matches both the memcg and blkcg of %current.
 270  * Must be called under rcu_read_lock() which protects the returend wb.
 271  * NULL if not found.
 272  */
 273 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
 274 {
 275         struct cgroup_subsys_state *memcg_css;
 276         struct bdi_writeback *wb;
 277
 278         memcg_css = task_css(current, memory_cgrp_id);
 279         if (!memcg_css->parent)
 280                 return &bdi->wb;
 281
 282         wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
 283
 284         /*
 285          * %current's blkcg equals the effective blkcg of its memcg.  No
 286          * need to use the relatively expensive cgroup_get_e_css().
 287          */
 288         if (likely(wb && wb->blkcg_css == task_css(current, blkio_cgrp_id)))
 289                 return wb;
 290         return NULL;
 291 }
 292
 293 /**
 294  * wb_get_create_current - get or create wb for %current on a bdi
 295  * @bdi: bdi of interest
 296  * @gfp: allocation mask
 297  *
 298  * Equivalent to wb_get_create() on %current's memcg.  This function is
 299  * called from a relatively hot path and optimizes the common cases using
 300  * wb_find_current().
 301  */
 302 static inline struct bdi_writeback *
 303 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
 304 {
 305         struct bdi_writeback *wb;
 306
 307         rcu_read_lock();
 308         wb = wb_find_current(bdi);
 309         if (wb && unlikely(!wb_tryget(wb)))
 310                 wb = NULL;
 311         rcu_read_unlock();
 312
 313         if (unlikely(!wb)) {
 314                 struct cgroup_subsys_state *memcg_css;
 315
 316                 memcg_css = task_get_css(current, memory_cgrp_id);
 317                 wb = wb_get_create(bdi, memcg_css, gfp);
 318                 css_put(memcg_css);
 319         }
 320         return wb;
 321 }
 322
 323 /**
 324  * inode_to_wb_is_valid - test whether an inode has a wb associated
 325  * @inode: inode of interest
 326  *
 327  * Returns %true if @inode has a wb associated.  May be called without any
 328  * locking.
 329  */
 330 static inline bool inode_to_wb_is_valid(struct inode *inode)
 331 {
 332         return inode->i_wb;
 333 }
 334
 335 /**
 336  * inode_to_wb - determine the wb of an inode
 337  * @inode: inode of interest
 338  *
 339  * Returns the wb @inode is currently associated with.  The caller must be
 340  * holding either @inode->i_lock, @inode->i_mapping->tree_lock, or the
 341  * associated wb's list_lock.
 342  */
 343 static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
 344 {
 345 #ifdef CONFIG_LOCKDEP
 346         WARN_ON_ONCE(debug_locks &&
 347                      (!lockdep_is_held(&inode->i_lock) &&
 348                       !lockdep_is_held(&inode->i_mapping->tree_lock) &&
 349                       !lockdep_is_held(&inode->i_wb->list_lock)));
 350 #endif
 351         return inode->i_wb;
 352 }
 353
 354 /**
 355  * unlocked_inode_to_wb_begin - begin unlocked inode wb access transaction
 356  * @inode: target inode
 357  * @lockedp: temp bool output param, to be passed to the end function
 358  *
 359  * The caller wants to access the wb associated with @inode but isn't
 360  * holding inode->i_lock, mapping->tree_lock or wb->list_lock.  This
 361  * function determines the wb associated with @inode and ensures that the
 362  * association doesn't change until the transaction is finished with
 363  * unlocked_inode_to_wb_end().
 364  *
 365  * The caller must call unlocked_inode_to_wb_end() with *@lockdep
 366  * afterwards and can't sleep during transaction.  IRQ may or may not be
 367  * disabled on return.
 368  */
 369 static inline struct bdi_writeback *
 370 unlocked_inode_to_wb_begin(struct inode *inode, bool *lockedp)
 371 {
 372         rcu_read_lock();
 373
 374         /*
 375          * Paired with store_release in inode_switch_wb_work_fn() and
 376          * ensures that we see the new wb if we see cleared I_WB_SWITCH.
 377          */
 378         *lockedp = smp_load_acquire(&inode->i_state) & I_WB_SWITCH;
 379
 380         if (unlikely(*lockedp))
 381                 spin_lock_irq(&inode->i_mapping->tree_lock);
 382
 383         /*
 384          * Protected by either !I_WB_SWITCH + rcu_read_lock() or tree_lock.
 385          * inode_to_wb() will bark.  Deref directly.
 386          */
 387         return inode->i_wb;
 388 }
 389
 390 /**
 391  * unlocked_inode_to_wb_end - end inode wb access transaction
 392  * @inode: target inode
 393  * @locked: *@lockedp from unlocked_inode_to_wb_begin()
 394  */
 395 static inline void unlocked_inode_to_wb_end(struct inode *inode, bool locked)
 396 {
 397         if (unlikely(locked))
 398                 spin_unlock_irq(&inode->i_mapping->tree_lock);
 399
 400         rcu_read_unlock();
 401 }
 402
 403 struct wb_iter {
 404         int                     start_blkcg_id;
 405         struct radix_tree_iter  tree_iter;
 406         void                    **slot;
 407 };
 408
 409 static inline struct bdi_writeback *__wb_iter_next(struct wb_iter *iter,
 410                                                    struct backing_dev_info *bdi)
 411 {
 412         struct radix_tree_iter *titer = &iter->tree_iter;
 413
 414         WARN_ON_ONCE(!rcu_read_lock_held());
 415
 416         if (iter->start_blkcg_id >= 0) {
 417                 iter->slot = radix_tree_iter_init(titer, iter->start_blkcg_id);
 418                 iter->start_blkcg_id = -1;
 419         } else {
 420                 iter->slot = radix_tree_next_slot(iter->slot, titer, 0);
 421         }
 422
 423         if (!iter->slot)
 424                 iter->slot = radix_tree_next_chunk(&bdi->cgwb_tree, titer, 0);
 425         if (iter->slot)
 426                 return *iter->slot;
 427         return NULL;
 428 }
 429
 430 static inline struct bdi_writeback *__wb_iter_init(struct wb_iter *iter,
 431                                                    struct backing_dev_info *bdi,
 432                                                    int start_blkcg_id)
 433 {
 434         iter->start_blkcg_id = start_blkcg_id;
 435
 436         if (start_blkcg_id)
 437                 return __wb_iter_next(iter, bdi);
 438         else
 439                 return &bdi->wb;
 440 }
 441
 442 /**
 443  * bdi_for_each_wb - walk all wb's of a bdi in ascending blkcg ID order
 444  * @wb_cur: cursor struct bdi_writeback pointer
 445  * @bdi: bdi to walk wb's of
 446  * @iter: pointer to struct wb_iter to be used as iteration buffer
 447  * @start_blkcg_id: blkcg ID to start iteration from
 448  *
 449  * Iterate @wb_cur through the wb's (bdi_writeback's) of @bdi in ascending
 450  * blkcg ID order starting from @start_blkcg_id.  @iter is struct wb_iter
 451  * to be used as temp storage during iteration.  rcu_read_lock() must be
 452  * held throughout iteration.
 453  */
 454 #define bdi_for_each_wb(wb_cur, bdi, iter, start_blkcg_id)              \
 455         for ((wb_cur) = __wb_iter_init(iter, bdi, start_blkcg_id);      \
 456              (wb_cur); (wb_cur) = __wb_iter_next(iter, bdi))
 457
 458 #else   /* CONFIG_CGROUP_WRITEBACK */
 459
 460 static inline bool inode_cgwb_enabled(struct inode *inode)
 461 {
 462         return false;
 463 }
 464
 465 static inline struct bdi_writeback_congested *
 466 wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp)
 467 {
 468         return bdi->wb.congested;
 469 }
 470
 471 static inline void wb_congested_put(struct bdi_writeback_congested *congested)
 472 {
 473 }
 474
 475 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
 476 {
 477         return &bdi->wb;
 478 }
 479
 480 static inline struct bdi_writeback *
 481 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
 482 {
 483         return &bdi->wb;
 484 }
 485
 486 static inline bool inode_to_wb_is_valid(struct inode *inode)
 487 {
 488         return true;
 489 }
 490
 491 static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
 492 {
 493         return &inode_to_bdi(inode)->wb;
 494 }
 495
 496 static inline struct bdi_writeback *
 497 unlocked_inode_to_wb_begin(struct inode *inode, bool *lockedp)
 498 {
 499         return inode_to_wb(inode);
 500 }
 501
 502 static inline void unlocked_inode_to_wb_end(struct inode *inode, bool locked)
 503 {
 504 }
 505
 506 static inline void wb_memcg_offline(struct mem_cgroup *memcg)
 507 {
 508 }
 509
 510 static inline void wb_blkcg_offline(struct blkcg *blkcg)
 511 {
 512 }
 513
 514 struct wb_iter {
 515         int             next_id;
 516 };
 517
 518 #define bdi_for_each_wb(wb_cur, bdi, iter, start_blkcg_id)              \
 519         for ((iter)->next_id = (start_blkcg_id);                        \
 520              ({ (wb_cur) = !(iter)->next_id++ ? &(bdi)->wb : NULL; }); )
 521
 522 static inline int inode_congested(struct inode *inode, int cong_bits)
 523 {
 524         return wb_congested(&inode_to_bdi(inode)->wb, cong_bits);
 525 }
 526
 527 #endif  /* CONFIG_CGROUP_WRITEBACK */
 528
 529 static inline int inode_read_congested(struct inode *inode)
 530 {
 531         return inode_congested(inode, 1 << WB_sync_congested);
 532 }
 533
 534 static inline int inode_write_congested(struct inode *inode)
 535 {
 536         return inode_congested(inode, 1 << WB_async_congested);
 537 }
 538
 539 static inline int inode_rw_congested(struct inode *inode)
 540 {
 541         return inode_congested(inode, (1 << WB_sync_congested) |
 542                                       (1 << WB_async_congested));
 543 }
 544
 545 static inline int bdi_congested(struct backing_dev_info *bdi, int cong_bits)
 546 {
 547         return wb_congested(&bdi->wb, cong_bits);
 548 }
 549
 550 static inline int bdi_read_congested(struct backing_dev_info *bdi)
 551 {
 552         return bdi_congested(bdi, 1 << WB_sync_congested);
 553 }
 554
 555 static inline int bdi_write_congested(struct backing_dev_info *bdi)
 556 {
 557         return bdi_congested(bdi, 1 << WB_async_congested);
 558 }
 559
 560 static inline int bdi_rw_congested(struct backing_dev_info *bdi)
 561 {
 562         return bdi_congested(bdi, (1 << WB_sync_congested) |
 563                                   (1 << WB_async_congested));
 564 }
 565
 566 #endif  /* _LINUX_BACKING_DEV_H */