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