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