include/linux/backing-dev.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 /*
   3  * include/linux/backing-dev.h
   4  *
   5  * low-level device information and state which is propagated up through
   6  * to high-level code.
   7  */
   8
   9 #ifndef _LINUX_BACKING_DEV_H
  10 #define _LINUX_BACKING_DEV_H
  11
  12 #include <linux/kernel.h>
  13 #include <linux/fs.h>
  14 #include <linux/sched.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/writeback.h>
  17 #include <linux/blk-cgroup.h>
  18 #include <linux/backing-dev-defs.h>
  19 #include <linux/slab.h>
  20
  21 static inline struct backing_dev_info *bdi_get(struct backing_dev_info *bdi)
  22 {
  23         kref_get(&bdi->refcnt);
  24         return bdi;
  25 }
  26
  27 struct backing_dev_info *bdi_get_by_id(u64 id);
  28 void bdi_put(struct backing_dev_info *bdi);
  29
  30 __printf(2, 3)
  31 int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...);
  32 __printf(2, 0)
  33 int bdi_register_va(struct backing_dev_info *bdi, const char *fmt,
  34                     va_list args);
  35 int bdi_register_owner(struct backing_dev_info *bdi, struct device *owner);
  36 void bdi_unregister(struct backing_dev_info *bdi);
  37
  38 struct backing_dev_info *bdi_alloc_node(gfp_t gfp_mask, int node_id);
  39 static inline struct backing_dev_info *bdi_alloc(gfp_t gfp_mask)
  40 {
  41         return bdi_alloc_node(gfp_mask, NUMA_NO_NODE);
  42 }
  43
  44 void wb_start_background_writeback(struct bdi_writeback *wb);
  45 void wb_workfn(struct work_struct *work);
  46 void wb_wakeup_delayed(struct bdi_writeback *wb);
  47
  48 void wb_wait_for_completion(struct wb_completion *done);
  49
  50 extern spinlock_t bdi_lock;
  51 extern struct list_head bdi_list;
  52
  53 extern struct workqueue_struct *bdi_wq;
  54 extern struct workqueue_struct *bdi_async_bio_wq;
  55
  56 static inline bool wb_has_dirty_io(struct bdi_writeback *wb)
  57 {
  58         return test_bit(WB_has_dirty_io, &wb->state);
  59 }
  60
  61 static inline bool bdi_has_dirty_io(struct backing_dev_info *bdi)
  62 {
  63         /*
  64          * @bdi->tot_write_bandwidth is guaranteed to be > 0 if there are
  65          * any dirty wbs.  See wb_update_write_bandwidth().
  66          */
  67         return atomic_long_read(&bdi->tot_write_bandwidth);
  68 }
  69
  70 static inline void __add_wb_stat(struct bdi_writeback *wb,
  71                                  enum wb_stat_item item, s64 amount)
  72 {
  73         percpu_counter_add_batch(&wb->stat[item], amount, WB_STAT_BATCH);
  74 }
  75
  76 static inline void inc_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
  77 {
  78         __add_wb_stat(wb, item, 1);
  79 }
  80
  81 static inline void dec_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
  82 {
  83         __add_wb_stat(wb, item, -1);
  84 }
  85
  86 static inline s64 wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
  87 {
  88         return percpu_counter_read_positive(&wb->stat[item]);
  89 }
  90
  91 static inline s64 wb_stat_sum(struct bdi_writeback *wb, enum wb_stat_item item)
  92 {
  93         return percpu_counter_sum_positive(&wb->stat[item]);
  94 }
  95
  96 extern void wb_writeout_inc(struct bdi_writeback *wb);
  97
  98 /*
  99  * maximal error of a stat counter.
 100  */
 101 static inline unsigned long wb_stat_error(void)
 102 {
 103 #ifdef CONFIG_SMP
 104         return nr_cpu_ids * WB_STAT_BATCH;
 105 #else
 106         return 1;
 107 #endif
 108 }
 109
 110 int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
 111 int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
 112
 113 /*
 114  * Flags in backing_dev_info::capability
 115  *
 116  * The first three flags control whether dirty pages will contribute to the
 117  * VM's accounting and whether writepages() should be called for dirty pages
 118  * (something that would not, for example, be appropriate for ramfs)
 119  *
 120  * WARNING: these flags are closely related and should not normally be
 121  * used separately.  The BDI_CAP_NO_ACCT_AND_WRITEBACK combines these
 122  * three flags into a single convenience macro.
 123  *
 124  * BDI_CAP_NO_ACCT_DIRTY:  Dirty pages shouldn't contribute to accounting
 125  * BDI_CAP_NO_WRITEBACK:   Don't write pages back
 126  * BDI_CAP_NO_ACCT_WB:     Don't automatically account writeback pages
 127  * BDI_CAP_STRICTLIMIT:    Keep number of dirty pages below bdi threshold.
 128  *
 129  * BDI_CAP_CGROUP_WRITEBACK: Supports cgroup-aware writeback.
 130  * BDI_CAP_SYNCHRONOUS_IO: Device is so fast that asynchronous IO would be
 131  *                         inefficient.
 132  */
 133 #define BDI_CAP_NO_ACCT_DIRTY   0x00000001
 134 #define BDI_CAP_NO_WRITEBACK    0x00000002
 135 #define BDI_CAP_NO_ACCT_WB      0x00000004
 136 #define BDI_CAP_STABLE_WRITES   0x00000008
 137 #define BDI_CAP_STRICTLIMIT     0x00000010
 138 #define BDI_CAP_CGROUP_WRITEBACK 0x00000020
 139 #define BDI_CAP_SYNCHRONOUS_IO  0x00000040
 140
 141 #define BDI_CAP_NO_ACCT_AND_WRITEBACK \
 142         (BDI_CAP_NO_WRITEBACK | BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_ACCT_WB)
 143
 144 extern struct backing_dev_info noop_backing_dev_info;
 145
 146 /**
 147  * writeback_in_progress - determine whether there is writeback in progress
 148  * @wb: bdi_writeback of interest
 149  *
 150  * Determine whether there is writeback waiting to be handled against a
 151  * bdi_writeback.
 152  */
 153 static inline bool writeback_in_progress(struct bdi_writeback *wb)
 154 {
 155         return test_bit(WB_writeback_running, &wb->state);
 156 }
 157
 158 static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
 159 {
 160         struct super_block *sb;
 161
 162         if (!inode)
 163                 return &noop_backing_dev_info;
 164
 165         sb = inode->i_sb;
 166 #ifdef CONFIG_BLOCK
 167         if (sb_is_blkdev_sb(sb))
 168                 return I_BDEV(inode)->bd_bdi;
 169 #endif
 170         return sb->s_bdi;
 171 }
 172
 173 static inline int wb_congested(struct bdi_writeback *wb, int cong_bits)
 174 {
 175         struct backing_dev_info *bdi = wb->bdi;
 176
 177         if (bdi->congested_fn)
 178                 return bdi->congested_fn(bdi->congested_data, cong_bits);
 179         return wb->congested->state & cong_bits;
 180 }
 181
 182 long congestion_wait(int sync, long timeout);
 183 long wait_iff_congested(int sync, long timeout);
 184
 185 static inline bool bdi_cap_synchronous_io(struct backing_dev_info *bdi)
 186 {
 187         return bdi->capabilities & BDI_CAP_SYNCHRONOUS_IO;
 188 }
 189
 190 static inline bool bdi_cap_stable_pages_required(struct backing_dev_info *bdi)
 191 {
 192         return bdi->capabilities & BDI_CAP_STABLE_WRITES;
 193 }
 194
 195 static inline bool bdi_cap_writeback_dirty(struct backing_dev_info *bdi)
 196 {
 197         return !(bdi->capabilities & BDI_CAP_NO_WRITEBACK);
 198 }
 199
 200 static inline bool bdi_cap_account_dirty(struct backing_dev_info *bdi)
 201 {
 202         return !(bdi->capabilities & BDI_CAP_NO_ACCT_DIRTY);
 203 }
 204
 205 static inline bool bdi_cap_account_writeback(struct backing_dev_info *bdi)
 206 {
 207         /* Paranoia: BDI_CAP_NO_WRITEBACK implies BDI_CAP_NO_ACCT_WB */
 208         return !(bdi->capabilities & (BDI_CAP_NO_ACCT_WB |
 209                                       BDI_CAP_NO_WRITEBACK));
 210 }
 211
 212 static inline bool mapping_cap_writeback_dirty(struct address_space *mapping)
 213 {
 214         return bdi_cap_writeback_dirty(inode_to_bdi(mapping->host));
 215 }
 216
 217 static inline bool mapping_cap_account_dirty(struct address_space *mapping)
 218 {
 219         return bdi_cap_account_dirty(inode_to_bdi(mapping->host));
 220 }
 221
 222 static inline int bdi_sched_wait(void *word)
 223 {
 224         schedule();
 225         return 0;
 226 }
 227
 228 #ifdef CONFIG_CGROUP_WRITEBACK
 229
 230 struct bdi_writeback_congested *
 231 wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp);
 232 void wb_congested_put(struct bdi_writeback_congested *congested);
 233 struct bdi_writeback *wb_get_lookup(struct backing_dev_info *bdi,
 234                                     struct cgroup_subsys_state *memcg_css);
 235 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
 236                                     struct cgroup_subsys_state *memcg_css,
 237                                     gfp_t gfp);
 238 void wb_memcg_offline(struct mem_cgroup *memcg);
 239 void wb_blkcg_offline(struct blkcg *blkcg);
 240 int inode_congested(struct inode *inode, int cong_bits);
 241
 242 /**
 243  * inode_cgwb_enabled - test whether cgroup writeback is enabled on an inode
 244  * @inode: inode of interest
 245  *
 246  * cgroup writeback requires support from both the bdi and filesystem.
 247  * Also, both memcg and iocg have to be on the default hierarchy.  Test
 248  * whether all conditions are met.
 249  *
 250  * Note that the test result may change dynamically on the same inode
 251  * depending on how memcg and iocg are configured.
 252  */
 253 static inline bool inode_cgwb_enabled(struct inode *inode)
 254 {
 255         struct backing_dev_info *bdi = inode_to_bdi(inode);
 256
 257         return cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
 258                 cgroup_subsys_on_dfl(io_cgrp_subsys) &&
 259                 bdi_cap_account_dirty(bdi) &&
 260                 (bdi->capabilities & BDI_CAP_CGROUP_WRITEBACK) &&
 261                 (inode->i_sb->s_iflags & SB_I_CGROUPWB);
 262 }
 263
 264 /**
 265  * wb_find_current - find wb for %current on a bdi
 266  * @bdi: bdi of interest
 267  *
 268  * Find the wb of @bdi which matches both the memcg and blkcg of %current.
 269  * Must be called under rcu_read_lock() which protects the returend wb.
 270  * NULL if not found.
 271  */
 272 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
 273 {
 274         struct cgroup_subsys_state *memcg_css;
 275         struct bdi_writeback *wb;
 276
 277         memcg_css = task_css(current, memory_cgrp_id);
 278         if (!memcg_css->parent)
 279                 return &bdi->wb;
 280
 281         wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
 282
 283         /*
 284          * %current's blkcg equals the effective blkcg of its memcg.  No
 285          * need to use the relatively expensive cgroup_get_e_css().
 286          */
 287         if (likely(wb && wb->blkcg_css == task_css(current, io_cgrp_id)))
 288                 return wb;
 289         return NULL;
 290 }
 291
 292 /**
 293  * wb_get_create_current - get or create wb for %current on a bdi
 294  * @bdi: bdi of interest
 295  * @gfp: allocation mask
 296  *
 297  * Equivalent to wb_get_create() on %current's memcg.  This function is
 298  * called from a relatively hot path and optimizes the common cases using
 299  * wb_find_current().
 300  */
 301 static inline struct bdi_writeback *
 302 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
 303 {
 304         struct bdi_writeback *wb;
 305
 306         rcu_read_lock();
 307         wb = wb_find_current(bdi);
 308         if (wb && unlikely(!wb_tryget(wb)))
 309                 wb = NULL;
 310         rcu_read_unlock();
 311
 312         if (unlikely(!wb)) {
 313                 struct cgroup_subsys_state *memcg_css;
 314
 315                 memcg_css = task_get_css(current, memory_cgrp_id);
 316                 wb = wb_get_create(bdi, memcg_css, gfp);
 317                 css_put(memcg_css);
 318         }
 319         return wb;
 320 }
 321
 322 /**
 323  * inode_to_wb_is_valid - test whether an inode has a wb associated
 324  * @inode: inode of interest
 325  *
 326  * Returns %true if @inode has a wb associated.  May be called without any
 327  * locking.
 328  */
 329 static inline bool inode_to_wb_is_valid(struct inode *inode)
 330 {
 331         return inode->i_wb;
 332 }
 333
 334 /**
 335  * inode_to_wb - determine the wb of an inode
 336  * @inode: inode of interest
 337  *
 338  * Returns the wb @inode is currently associated with.  The caller must be
 339  * holding either @inode->i_lock, the i_pages lock, or the
 340  * associated wb's list_lock.
 341  */
 342 static inline struct bdi_writeback *inode_to_wb(const struct inode *inode)
 343 {
 344 #ifdef CONFIG_LOCKDEP
 345         WARN_ON_ONCE(debug_locks &&
 346                      (!lockdep_is_held(&inode->i_lock) &&
 347                       !lockdep_is_held(&inode->i_mapping->i_pages.xa_lock) &&
 348                       !lockdep_is_held(&inode->i_wb->list_lock)));
 349 #endif
 350         return inode->i_wb;
 351 }
 352
 353 /**
 354  * unlocked_inode_to_wb_begin - begin unlocked inode wb access transaction
 355  * @inode: target inode
 356  * @cookie: output param, to be passed to the end function
 357  *
 358  * The caller wants to access the wb associated with @inode but isn't
 359  * holding inode->i_lock, the i_pages lock or wb->list_lock.  This
 360  * function determines the wb associated with @inode and ensures that the
 361  * association doesn't change until the transaction is finished with
 362  * unlocked_inode_to_wb_end().
 363  *
 364  * The caller must call unlocked_inode_to_wb_end() with *@cookie afterwards and
 365  * can't sleep during the transaction.  IRQs may or may not be disabled on
 366  * return.
 367  */
 368 static inline struct bdi_writeback *
 369 unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie)
 370 {
 371         rcu_read_lock();
 372
 373         /*
 374          * Paired with store_release in inode_switch_wbs_work_fn() and
 375          * ensures that we see the new wb if we see cleared I_WB_SWITCH.
 376          */
 377         cookie->locked = smp_load_acquire(&inode->i_state) & I_WB_SWITCH;
 378
 379         if (unlikely(cookie->locked))
 380                 xa_lock_irqsave(&inode->i_mapping->i_pages, cookie->flags);
 381
 382         /*
 383          * Protected by either !I_WB_SWITCH + rcu_read_lock() or the i_pages
 384          * lock.  inode_to_wb() will bark.  Deref directly.
 385          */
 386         return inode->i_wb;
 387 }
 388
 389 /**
 390  * unlocked_inode_to_wb_end - end inode wb access transaction
 391  * @inode: target inode
 392  * @cookie: @cookie from unlocked_inode_to_wb_begin()
 393  */
 394 static inline void unlocked_inode_to_wb_end(struct inode *inode,
 395                                             struct wb_lock_cookie *cookie)
 396 {
 397         if (unlikely(cookie->locked))
 398                 xa_unlock_irqrestore(&inode->i_mapping->i_pages, cookie->flags);
 399
 400         rcu_read_unlock();
 401 }
 402
 403 #else   /* CONFIG_CGROUP_WRITEBACK */
 404
 405 static inline bool inode_cgwb_enabled(struct inode *inode)
 406 {
 407         return false;
 408 }
 409
 410 static inline struct bdi_writeback_congested *
 411 wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp)
 412 {
 413         refcount_inc(&bdi->wb_congested->refcnt);
 414         return bdi->wb_congested;
 415 }
 416
 417 static inline void wb_congested_put(struct bdi_writeback_congested *congested)
 418 {
 419         if (refcount_dec_and_test(&congested->refcnt))
 420                 kfree(congested);
 421 }
 422
 423 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
 424 {
 425         return &bdi->wb;
 426 }
 427
 428 static inline struct bdi_writeback *
 429 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
 430 {
 431         return &bdi->wb;
 432 }
 433
 434 static inline bool inode_to_wb_is_valid(struct inode *inode)
 435 {
 436         return true;
 437 }
 438
 439 static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
 440 {
 441         return &inode_to_bdi(inode)->wb;
 442 }
 443
 444 static inline struct bdi_writeback *
 445 unlocked_inode_to_wb_begin(struct inode *inode, struct wb_lock_cookie *cookie)
 446 {
 447         return inode_to_wb(inode);
 448 }
 449
 450 static inline void unlocked_inode_to_wb_end(struct inode *inode,
 451                                             struct wb_lock_cookie *cookie)
 452 {
 453 }
 454
 455 static inline void wb_memcg_offline(struct mem_cgroup *memcg)
 456 {
 457 }
 458
 459 static inline void wb_blkcg_offline(struct blkcg *blkcg)
 460 {
 461 }
 462
 463 static inline int inode_congested(struct inode *inode, int cong_bits)
 464 {
 465         return wb_congested(&inode_to_bdi(inode)->wb, cong_bits);
 466 }
 467
 468 #endif  /* CONFIG_CGROUP_WRITEBACK */
 469
 470 static inline int inode_read_congested(struct inode *inode)
 471 {
 472         return inode_congested(inode, 1 << WB_sync_congested);
 473 }
 474
 475 static inline int inode_write_congested(struct inode *inode)
 476 {
 477         return inode_congested(inode, 1 << WB_async_congested);
 478 }
 479
 480 static inline int inode_rw_congested(struct inode *inode)
 481 {
 482         return inode_congested(inode, (1 << WB_sync_congested) |
 483                                       (1 << WB_async_congested));
 484 }
 485
 486 static inline int bdi_congested(struct backing_dev_info *bdi, int cong_bits)
 487 {
 488         return wb_congested(&bdi->wb, cong_bits);
 489 }
 490
 491 static inline int bdi_read_congested(struct backing_dev_info *bdi)
 492 {
 493         return bdi_congested(bdi, 1 << WB_sync_congested);
 494 }
 495
 496 static inline int bdi_write_congested(struct backing_dev_info *bdi)
 497 {
 498         return bdi_congested(bdi, 1 << WB_async_congested);
 499 }
 500
 501 static inline int bdi_rw_congested(struct backing_dev_info *bdi)
 502 {
 503         return bdi_congested(bdi, (1 << WB_sync_congested) |
 504                                   (1 << WB_async_congested));
 505 }
 506
 507 #endif  /* _LINUX_BACKING_DEV_H */