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