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 #include <linux/slab.h>
  19
  20 int __must_check bdi_init(struct backing_dev_info *bdi);
  21 void bdi_destroy(struct backing_dev_info *bdi);
  22
  23 __printf(3, 4)
  24 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
  25                 const char *fmt, ...);
  26 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev);
  27 int __must_check bdi_setup_and_register(struct backing_dev_info *, char *);
  28 void wb_start_writeback(struct bdi_writeback *wb, long nr_pages,
  29                         bool range_cyclic, enum wb_reason reason);
  30 void wb_start_background_writeback(struct bdi_writeback *wb);
  31 void wb_workfn(struct work_struct *work);
  32 void wb_wakeup_delayed(struct bdi_writeback *wb);
  33
  34 extern spinlock_t bdi_lock;
  35 extern struct list_head bdi_list;
  36
  37 extern struct workqueue_struct *bdi_wq;
  38
  39 static inline bool wb_has_dirty_io(struct bdi_writeback *wb)
  40 {
  41         return test_bit(WB_has_dirty_io, &wb->state);
  42 }
  43
  44 static inline bool bdi_has_dirty_io(struct backing_dev_info *bdi)
  45 {
  46         /*
  47          * @bdi->tot_write_bandwidth is guaranteed to be > 0 if there are
  48          * any dirty wbs.  See wb_update_write_bandwidth().
  49          */
  50         return atomic_long_read(&bdi->tot_write_bandwidth);
  51 }
  52
  53 static inline void __add_wb_stat(struct bdi_writeback *wb,
  54                                  enum wb_stat_item item, s64 amount)
  55 {
  56         __percpu_counter_add(&wb->stat[item], amount, WB_STAT_BATCH);
  57 }
  58
  59 static inline void __inc_wb_stat(struct bdi_writeback *wb,
  60                                  enum wb_stat_item item)
  61 {
  62         __add_wb_stat(wb, item, 1);
  63 }
  64
  65 static inline void inc_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
  66 {
  67         unsigned long flags;
  68
  69         local_irq_save(flags);
  70         __inc_wb_stat(wb, item);
  71         local_irq_restore(flags);
  72 }
  73
  74 static inline void __dec_wb_stat(struct bdi_writeback *wb,
  75                                  enum wb_stat_item item)
  76 {
  77         __add_wb_stat(wb, item, -1);
  78 }
  79
  80 static inline void dec_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
  81 {
  82         unsigned long flags;
  83
  84         local_irq_save(flags);
  85         __dec_wb_stat(wb, item);
  86         local_irq_restore(flags);
  87 }
  88
  89 static inline s64 wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
  90 {
  91         return percpu_counter_read_positive(&wb->stat[item]);
  92 }
  93
  94 static inline s64 __wb_stat_sum(struct bdi_writeback *wb,
  95                                 enum wb_stat_item item)
  96 {
  97         return percpu_counter_sum_positive(&wb->stat[item]);
  98 }
  99
 100 static inline s64 wb_stat_sum(struct bdi_writeback *wb, enum wb_stat_item item)
 101 {
 102         s64 sum;
 103         unsigned long flags;
 104
 105         local_irq_save(flags);
 106         sum = __wb_stat_sum(wb, item);
 107         local_irq_restore(flags);
 108
 109         return sum;
 110 }
 111
 112 extern void wb_writeout_inc(struct bdi_writeback *wb);
 113
 114 /*
 115  * maximal error of a stat counter.
 116  */
 117 static inline unsigned long wb_stat_error(struct bdi_writeback *wb)
 118 {
 119 #ifdef CONFIG_SMP
 120         return nr_cpu_ids * WB_STAT_BATCH;
 121 #else
 122         return 1;
 123 #endif
 124 }
 125
 126 int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
 127 int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
 128
 129 /*
 130  * Flags in backing_dev_info::capability
 131  *
 132  * The first three flags control whether dirty pages will contribute to the
 133  * VM's accounting and whether writepages() should be called for dirty pages
 134  * (something that would not, for example, be appropriate for ramfs)
 135  *
 136  * WARNING: these flags are closely related and should not normally be
 137  * used separately.  The BDI_CAP_NO_ACCT_AND_WRITEBACK combines these
 138  * three flags into a single convenience macro.
 139  *
 140  * BDI_CAP_NO_ACCT_DIRTY:  Dirty pages shouldn't contribute to accounting
 141  * BDI_CAP_NO_WRITEBACK:   Don't write pages back
 142  * BDI_CAP_NO_ACCT_WB:     Don't automatically account writeback pages
 143  * BDI_CAP_STRICTLIMIT:    Keep number of dirty pages below bdi threshold.
 144  *
 145  * BDI_CAP_CGROUP_WRITEBACK: Supports cgroup-aware writeback.
 146  */
 147 #define BDI_CAP_NO_ACCT_DIRTY   0x00000001
 148 #define BDI_CAP_NO_WRITEBACK    0x00000002
 149 #define BDI_CAP_NO_ACCT_WB      0x00000004
 150 #define BDI_CAP_STABLE_WRITES   0x00000008
 151 #define BDI_CAP_STRICTLIMIT     0x00000010
 152 #define BDI_CAP_CGROUP_WRITEBACK 0x00000020
 153
 154 #define BDI_CAP_NO_ACCT_AND_WRITEBACK \
 155         (BDI_CAP_NO_WRITEBACK | BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_ACCT_WB)
 156
 157 extern struct backing_dev_info noop_backing_dev_info;
 158
 159 /**
 160  * writeback_in_progress - determine whether there is writeback in progress
 161  * @wb: bdi_writeback of interest
 162  *
 163  * Determine whether there is writeback waiting to be handled against a
 164  * bdi_writeback.
 165  */
 166 static inline bool writeback_in_progress(struct bdi_writeback *wb)
 167 {
 168         return test_bit(WB_writeback_running, &wb->state);
 169 }
 170
 171 static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
 172 {
 173         struct super_block *sb;
 174
 175         if (!inode)
 176                 return &noop_backing_dev_info;
 177
 178         sb = inode->i_sb;
 179 #ifdef CONFIG_BLOCK
 180         if (sb_is_blkdev_sb(sb))
 181                 return blk_get_backing_dev_info(I_BDEV(inode));
 182 #endif
 183         return sb->s_bdi;
 184 }
 185
 186 static inline int wb_congested(struct bdi_writeback *wb, int cong_bits)
 187 {
 188         struct backing_dev_info *bdi = wb->bdi;
 189
 190         if (bdi->congested_fn)
 191                 return bdi->congested_fn(bdi->congested_data, cong_bits);
 192         return wb->congested->state & cong_bits;
 193 }
 194
 195 long congestion_wait(int sync, long timeout);
 196 long wait_iff_congested(struct zone *zone, int sync, long timeout);
 197 int pdflush_proc_obsolete(struct ctl_table *table, int write,
 198                 void __user *buffer, size_t *lenp, loff_t *ppos);
 199
 200 static inline bool bdi_cap_stable_pages_required(struct backing_dev_info *bdi)
 201 {
 202         return bdi->capabilities & BDI_CAP_STABLE_WRITES;
 203 }
 204
 205 static inline bool bdi_cap_writeback_dirty(struct backing_dev_info *bdi)
 206 {
 207         return !(bdi->capabilities & BDI_CAP_NO_WRITEBACK);
 208 }
 209
 210 static inline bool bdi_cap_account_dirty(struct backing_dev_info *bdi)
 211 {
 212         return !(bdi->capabilities & BDI_CAP_NO_ACCT_DIRTY);
 213 }
 214
 215 static inline bool bdi_cap_account_writeback(struct backing_dev_info *bdi)
 216 {
 217         /* Paranoia: BDI_CAP_NO_WRITEBACK implies BDI_CAP_NO_ACCT_WB */
 218         return !(bdi->capabilities & (BDI_CAP_NO_ACCT_WB |
 219                                       BDI_CAP_NO_WRITEBACK));
 220 }
 221
 222 static inline bool mapping_cap_writeback_dirty(struct address_space *mapping)
 223 {
 224         return bdi_cap_writeback_dirty(inode_to_bdi(mapping->host));
 225 }
 226
 227 static inline bool mapping_cap_account_dirty(struct address_space *mapping)
 228 {
 229         return bdi_cap_account_dirty(inode_to_bdi(mapping->host));
 230 }
 231
 232 static inline int bdi_sched_wait(void *word)
 233 {
 234         schedule();
 235         return 0;
 236 }
 237
 238 #ifdef CONFIG_CGROUP_WRITEBACK
 239
 240 struct bdi_writeback_congested *
 241 wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp);
 242 void wb_congested_put(struct bdi_writeback_congested *congested);
 243 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
 244                                     struct cgroup_subsys_state *memcg_css,
 245                                     gfp_t gfp);
 246 void wb_memcg_offline(struct mem_cgroup *memcg);
 247 void wb_blkcg_offline(struct blkcg *blkcg);
 248 int inode_congested(struct inode *inode, int cong_bits);
 249
 250 /**
 251  * inode_cgwb_enabled - test whether cgroup writeback is enabled on an inode
 252  * @inode: inode of interest
 253  *
 254  * cgroup writeback requires support from both the bdi and filesystem.
 255  * Test whether @inode has both.
 256  */
 257 static inline bool inode_cgwb_enabled(struct inode *inode)
 258 {
 259         struct backing_dev_info *bdi = inode_to_bdi(inode);
 260
 261         return bdi_cap_account_dirty(bdi) &&
 262                 (bdi->capabilities & BDI_CAP_CGROUP_WRITEBACK) &&
 263                 (inode->i_sb->s_iflags & SB_I_CGROUPWB);
 264 }
 265
 266 /**
 267  * wb_find_current - find wb for %current on a bdi
 268  * @bdi: bdi of interest
 269  *
 270  * Find the wb of @bdi which matches both the memcg and blkcg of %current.
 271  * Must be called under rcu_read_lock() which protects the returend wb.
 272  * NULL if not found.
 273  */
 274 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
 275 {
 276         struct cgroup_subsys_state *memcg_css;
 277         struct bdi_writeback *wb;
 278
 279         memcg_css = task_css(current, memory_cgrp_id);
 280         if (!memcg_css->parent)
 281                 return &bdi->wb;
 282
 283         wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
 284
 285         /*
 286          * %current's blkcg equals the effective blkcg of its memcg.  No
 287          * need to use the relatively expensive cgroup_get_e_css().
 288          */
 289         if (likely(wb && wb->blkcg_css == task_css(current, blkio_cgrp_id)))
 290                 return wb;
 291         return NULL;
 292 }
 293
 294 /**
 295  * wb_get_create_current - get or create wb for %current on a bdi
 296  * @bdi: bdi of interest
 297  * @gfp: allocation mask
 298  *
 299  * Equivalent to wb_get_create() on %current's memcg.  This function is
 300  * called from a relatively hot path and optimizes the common cases using
 301  * wb_find_current().
 302  */
 303 static inline struct bdi_writeback *
 304 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
 305 {
 306         struct bdi_writeback *wb;
 307
 308         rcu_read_lock();
 309         wb = wb_find_current(bdi);
 310         if (wb && unlikely(!wb_tryget(wb)))
 311                 wb = NULL;
 312         rcu_read_unlock();
 313
 314         if (unlikely(!wb)) {
 315                 struct cgroup_subsys_state *memcg_css;
 316
 317                 memcg_css = task_get_css(current, memory_cgrp_id);
 318                 wb = wb_get_create(bdi, memcg_css, gfp);
 319                 css_put(memcg_css);
 320         }
 321         return wb;
 322 }
 323
 324 /**
 325  * inode_to_wb_is_valid - test whether an inode has a wb associated
 326  * @inode: inode of interest
 327  *
 328  * Returns %true if @inode has a wb associated.  May be called without any
 329  * locking.
 330  */
 331 static inline bool inode_to_wb_is_valid(struct inode *inode)
 332 {
 333         return inode->i_wb;
 334 }
 335
 336 /**
 337  * inode_to_wb - determine the wb of an inode
 338  * @inode: inode of interest
 339  *
 340  * Returns the wb @inode is currently associated with.  The caller must be
 341  * holding either @inode->i_lock, @inode->i_mapping->tree_lock, or the
 342  * associated wb's list_lock.
 343  */
 344 static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
 345 {
 346 #ifdef CONFIG_LOCKDEP
 347         WARN_ON_ONCE(debug_locks &&
 348                      (!lockdep_is_held(&inode->i_lock) &&
 349                       !lockdep_is_held(&inode->i_mapping->tree_lock) &&
 350                       !lockdep_is_held(&inode->i_wb->list_lock)));
 351 #endif
 352         return inode->i_wb;
 353 }
 354
 355 /**
 356  * unlocked_inode_to_wb_begin - begin unlocked inode wb access transaction
 357  * @inode: target inode
 358  * @lockedp: temp bool output param, to be passed to the end function
 359  *
 360  * The caller wants to access the wb associated with @inode but isn't
 361  * holding inode->i_lock, mapping->tree_lock or wb->list_lock.  This
 362  * function determines the wb associated with @inode and ensures that the
 363  * association doesn't change until the transaction is finished with
 364  * unlocked_inode_to_wb_end().
 365  *
 366  * The caller must call unlocked_inode_to_wb_end() with *@lockdep
 367  * afterwards and can't sleep during transaction.  IRQ may or may not be
 368  * disabled on return.
 369  */
 370 static inline struct bdi_writeback *
 371 unlocked_inode_to_wb_begin(struct inode *inode, bool *lockedp)
 372 {
 373         rcu_read_lock();
 374
 375         /*
 376          * Paired with store_release in inode_switch_wb_work_fn() and
 377          * ensures that we see the new wb if we see cleared I_WB_SWITCH.
 378          */
 379         *lockedp = smp_load_acquire(&inode->i_state) & I_WB_SWITCH;
 380
 381         if (unlikely(*lockedp))
 382                 spin_lock_irq(&inode->i_mapping->tree_lock);
 383
 384         /*
 385          * Protected by either !I_WB_SWITCH + rcu_read_lock() or tree_lock.
 386          * inode_to_wb() will bark.  Deref directly.
 387          */
 388         return inode->i_wb;
 389 }
 390
 391 /**
 392  * unlocked_inode_to_wb_end - end inode wb access transaction
 393  * @inode: target inode
 394  * @locked: *@lockedp from unlocked_inode_to_wb_begin()
 395  */
 396 static inline void unlocked_inode_to_wb_end(struct inode *inode, bool locked)
 397 {
 398         if (unlikely(locked))
 399                 spin_unlock_irq(&inode->i_mapping->tree_lock);
 400
 401         rcu_read_unlock();
 402 }
 403
 404 struct wb_iter {
 405         int                     start_blkcg_id;
 406         struct radix_tree_iter  tree_iter;
 407         void                    **slot;
 408 };
 409
 410 static inline struct bdi_writeback *__wb_iter_next(struct wb_iter *iter,
 411                                                    struct backing_dev_info *bdi)
 412 {
 413         struct radix_tree_iter *titer = &iter->tree_iter;
 414
 415         WARN_ON_ONCE(!rcu_read_lock_held());
 416
 417         if (iter->start_blkcg_id >= 0) {
 418                 iter->slot = radix_tree_iter_init(titer, iter->start_blkcg_id);
 419                 iter->start_blkcg_id = -1;
 420         } else {
 421                 iter->slot = radix_tree_next_slot(iter->slot, titer, 0);
 422         }
 423
 424         if (!iter->slot)
 425                 iter->slot = radix_tree_next_chunk(&bdi->cgwb_tree, titer, 0);
 426         if (iter->slot)
 427                 return *iter->slot;
 428         return NULL;
 429 }
 430
 431 static inline struct bdi_writeback *__wb_iter_init(struct wb_iter *iter,
 432                                                    struct backing_dev_info *bdi,
 433                                                    int start_blkcg_id)
 434 {
 435         iter->start_blkcg_id = start_blkcg_id;
 436
 437         if (start_blkcg_id)
 438                 return __wb_iter_next(iter, bdi);
 439         else
 440                 return &bdi->wb;
 441 }
 442
 443 /**
 444  * bdi_for_each_wb - walk all wb's of a bdi in ascending blkcg ID order
 445  * @wb_cur: cursor struct bdi_writeback pointer
 446  * @bdi: bdi to walk wb's of
 447  * @iter: pointer to struct wb_iter to be used as iteration buffer
 448  * @start_blkcg_id: blkcg ID to start iteration from
 449  *
 450  * Iterate @wb_cur through the wb's (bdi_writeback's) of @bdi in ascending
 451  * blkcg ID order starting from @start_blkcg_id.  @iter is struct wb_iter
 452  * to be used as temp storage during iteration.  rcu_read_lock() must be
 453  * held throughout iteration.
 454  */
 455 #define bdi_for_each_wb(wb_cur, bdi, iter, start_blkcg_id)              \
 456         for ((wb_cur) = __wb_iter_init(iter, bdi, start_blkcg_id);      \
 457              (wb_cur); (wb_cur) = __wb_iter_next(iter, bdi))
 458
 459 #else   /* CONFIG_CGROUP_WRITEBACK */
 460
 461 static inline bool inode_cgwb_enabled(struct inode *inode)
 462 {
 463         return false;
 464 }
 465
 466 static inline struct bdi_writeback_congested *
 467 wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp)
 468 {
 469         atomic_inc(&bdi->wb_congested->refcnt);
 470         return bdi->wb_congested;
 471 }
 472
 473 static inline void wb_congested_put(struct bdi_writeback_congested *congested)
 474 {
 475         if (atomic_dec_and_test(&congested->refcnt))
 476                 kfree(congested);
 477 }
 478
 479 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
 480 {
 481         return &bdi->wb;
 482 }
 483
 484 static inline struct bdi_writeback *
 485 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
 486 {
 487         return &bdi->wb;
 488 }
 489
 490 static inline bool inode_to_wb_is_valid(struct inode *inode)
 491 {
 492         return true;
 493 }
 494
 495 static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
 496 {
 497         return &inode_to_bdi(inode)->wb;
 498 }
 499
 500 static inline struct bdi_writeback *
 501 unlocked_inode_to_wb_begin(struct inode *inode, bool *lockedp)
 502 {
 503         return inode_to_wb(inode);
 504 }
 505
 506 static inline void unlocked_inode_to_wb_end(struct inode *inode, bool locked)
 507 {
 508 }
 509
 510 static inline void wb_memcg_offline(struct mem_cgroup *memcg)
 511 {
 512 }
 513
 514 static inline void wb_blkcg_offline(struct blkcg *blkcg)
 515 {
 516 }
 517
 518 struct wb_iter {
 519         int             next_id;
 520 };
 521
 522 #define bdi_for_each_wb(wb_cur, bdi, iter, start_blkcg_id)              \
 523         for ((iter)->next_id = (start_blkcg_id);                        \
 524              ({ (wb_cur) = !(iter)->next_id++ ? &(bdi)->wb : NULL; }); )
 525
 526 static inline int inode_congested(struct inode *inode, int cong_bits)
 527 {
 528         return wb_congested(&inode_to_bdi(inode)->wb, cong_bits);
 529 }
 530
 531 #endif  /* CONFIG_CGROUP_WRITEBACK */
 532
 533 static inline int inode_read_congested(struct inode *inode)
 534 {
 535         return inode_congested(inode, 1 << WB_sync_congested);
 536 }
 537
 538 static inline int inode_write_congested(struct inode *inode)
 539 {
 540         return inode_congested(inode, 1 << WB_async_congested);
 541 }
 542
 543 static inline int inode_rw_congested(struct inode *inode)
 544 {
 545         return inode_congested(inode, (1 << WB_sync_congested) |
 546                                       (1 << WB_async_congested));
 547 }
 548
 549 static inline int bdi_congested(struct backing_dev_info *bdi, int cong_bits)
 550 {
 551         return wb_congested(&bdi->wb, cong_bits);
 552 }
 553
 554 static inline int bdi_read_congested(struct backing_dev_info *bdi)
 555 {
 556         return bdi_congested(bdi, 1 << WB_sync_congested);
 557 }
 558
 559 static inline int bdi_write_congested(struct backing_dev_info *bdi)
 560 {
 561         return bdi_congested(bdi, 1 << WB_async_congested);
 562 }
 563
 564 static inline int bdi_rw_congested(struct backing_dev_info *bdi)
 565 {
 566         return bdi_congested(bdi, (1 << WB_sync_congested) |
 567                                   (1 << WB_async_congested));
 568 }
 569
 570 #endif  /* _LINUX_BACKING_DEV_H */