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