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