2 * include/linux/backing-dev.h
4 * low-level device information and state which is propagated up through
8 #ifndef _LINUX_BACKING_DEV_H
9 #define _LINUX_BACKING_DEV_H
11 #include <linux/kernel.h>
13 #include <linux/sched.h>
14 #include <linux/blkdev.h>
15 #include <linux/writeback.h>
16 #include <linux/memcontrol.h>
17 #include <linux/blk-cgroup.h>
18 #include <linux/backing-dev-defs.h>
19 #include <linux/slab.h>
21 int __must_check bdi_init(struct backing_dev_info *bdi);
22 void bdi_destroy(struct backing_dev_info *bdi);
25 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
26 const char *fmt, ...);
27 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev);
28 int __must_check bdi_setup_and_register(struct backing_dev_info *, char *);
29 void wb_start_writeback(struct bdi_writeback *wb, long nr_pages,
30 bool range_cyclic, enum wb_reason reason);
31 void wb_start_background_writeback(struct bdi_writeback *wb);
32 void wb_workfn(struct work_struct *work);
33 void wb_wakeup_delayed(struct bdi_writeback *wb);
35 extern spinlock_t bdi_lock;
36 extern struct list_head bdi_list;
38 extern struct workqueue_struct *bdi_wq;
40 static inline bool wb_has_dirty_io(struct bdi_writeback *wb)
42 return test_bit(WB_has_dirty_io, &wb->state);
45 static inline bool bdi_has_dirty_io(struct backing_dev_info *bdi)
48 * @bdi->tot_write_bandwidth is guaranteed to be > 0 if there are
49 * any dirty wbs. See wb_update_write_bandwidth().
51 return atomic_long_read(&bdi->tot_write_bandwidth);
54 static inline void __add_wb_stat(struct bdi_writeback *wb,
55 enum wb_stat_item item, s64 amount)
57 __percpu_counter_add(&wb->stat[item], amount, WB_STAT_BATCH);
60 static inline void __inc_wb_stat(struct bdi_writeback *wb,
61 enum wb_stat_item item)
63 __add_wb_stat(wb, item, 1);
66 static inline void inc_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
70 local_irq_save(flags);
71 __inc_wb_stat(wb, item);
72 local_irq_restore(flags);
75 static inline void __dec_wb_stat(struct bdi_writeback *wb,
76 enum wb_stat_item item)
78 __add_wb_stat(wb, item, -1);
81 static inline void dec_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
85 local_irq_save(flags);
86 __dec_wb_stat(wb, item);
87 local_irq_restore(flags);
90 static inline s64 wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
92 return percpu_counter_read_positive(&wb->stat[item]);
95 static inline s64 __wb_stat_sum(struct bdi_writeback *wb,
96 enum wb_stat_item item)
98 return percpu_counter_sum_positive(&wb->stat[item]);
101 static inline s64 wb_stat_sum(struct bdi_writeback *wb, enum wb_stat_item item)
106 local_irq_save(flags);
107 sum = __wb_stat_sum(wb, item);
108 local_irq_restore(flags);
113 extern void wb_writeout_inc(struct bdi_writeback *wb);
116 * maximal error of a stat counter.
118 static inline unsigned long wb_stat_error(struct bdi_writeback *wb)
121 return nr_cpu_ids * WB_STAT_BATCH;
127 int bdi_set_min_ratio(struct backing_dev_info *bdi, unsigned int min_ratio);
128 int bdi_set_max_ratio(struct backing_dev_info *bdi, unsigned int max_ratio);
131 * Flags in backing_dev_info::capability
133 * The first three flags control whether dirty pages will contribute to the
134 * VM's accounting and whether writepages() should be called for dirty pages
135 * (something that would not, for example, be appropriate for ramfs)
137 * WARNING: these flags are closely related and should not normally be
138 * used separately. The BDI_CAP_NO_ACCT_AND_WRITEBACK combines these
139 * three flags into a single convenience macro.
141 * BDI_CAP_NO_ACCT_DIRTY: Dirty pages shouldn't contribute to accounting
142 * BDI_CAP_NO_WRITEBACK: Don't write pages back
143 * BDI_CAP_NO_ACCT_WB: Don't automatically account writeback pages
144 * BDI_CAP_STRICTLIMIT: Keep number of dirty pages below bdi threshold.
146 * BDI_CAP_CGROUP_WRITEBACK: Supports cgroup-aware writeback.
148 #define BDI_CAP_NO_ACCT_DIRTY 0x00000001
149 #define BDI_CAP_NO_WRITEBACK 0x00000002
150 #define BDI_CAP_NO_ACCT_WB 0x00000004
151 #define BDI_CAP_STABLE_WRITES 0x00000008
152 #define BDI_CAP_STRICTLIMIT 0x00000010
153 #define BDI_CAP_CGROUP_WRITEBACK 0x00000020
155 #define BDI_CAP_NO_ACCT_AND_WRITEBACK \
156 (BDI_CAP_NO_WRITEBACK | BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_ACCT_WB)
158 extern struct backing_dev_info noop_backing_dev_info;
161 * writeback_in_progress - determine whether there is writeback in progress
162 * @wb: bdi_writeback of interest
164 * Determine whether there is writeback waiting to be handled against a
167 static inline bool writeback_in_progress(struct bdi_writeback *wb)
169 return test_bit(WB_writeback_running, &wb->state);
172 static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
174 struct super_block *sb;
177 return &noop_backing_dev_info;
181 if (sb_is_blkdev_sb(sb))
182 return blk_get_backing_dev_info(I_BDEV(inode));
187 static inline int wb_congested(struct bdi_writeback *wb, int cong_bits)
189 struct backing_dev_info *bdi = wb->bdi;
191 if (bdi->congested_fn)
192 return bdi->congested_fn(bdi->congested_data, cong_bits);
193 return wb->congested->state & cong_bits;
196 long congestion_wait(int sync, long timeout);
197 long wait_iff_congested(struct zone *zone, int sync, long timeout);
198 int pdflush_proc_obsolete(struct ctl_table *table, int write,
199 void __user *buffer, size_t *lenp, loff_t *ppos);
201 static inline bool bdi_cap_stable_pages_required(struct backing_dev_info *bdi)
203 return bdi->capabilities & BDI_CAP_STABLE_WRITES;
206 static inline bool bdi_cap_writeback_dirty(struct backing_dev_info *bdi)
208 return !(bdi->capabilities & BDI_CAP_NO_WRITEBACK);
211 static inline bool bdi_cap_account_dirty(struct backing_dev_info *bdi)
213 return !(bdi->capabilities & BDI_CAP_NO_ACCT_DIRTY);
216 static inline bool bdi_cap_account_writeback(struct backing_dev_info *bdi)
218 /* Paranoia: BDI_CAP_NO_WRITEBACK implies BDI_CAP_NO_ACCT_WB */
219 return !(bdi->capabilities & (BDI_CAP_NO_ACCT_WB |
220 BDI_CAP_NO_WRITEBACK));
223 static inline bool mapping_cap_writeback_dirty(struct address_space *mapping)
225 return bdi_cap_writeback_dirty(inode_to_bdi(mapping->host));
228 static inline bool mapping_cap_account_dirty(struct address_space *mapping)
230 return bdi_cap_account_dirty(inode_to_bdi(mapping->host));
233 static inline int bdi_sched_wait(void *word)
239 #ifdef CONFIG_CGROUP_WRITEBACK
241 struct bdi_writeback_congested *
242 wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp);
243 void wb_congested_put(struct bdi_writeback_congested *congested);
244 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
245 struct cgroup_subsys_state *memcg_css,
247 void wb_memcg_offline(struct mem_cgroup *memcg);
248 void wb_blkcg_offline(struct blkcg *blkcg);
249 int inode_congested(struct inode *inode, int cong_bits);
252 * inode_cgwb_enabled - test whether cgroup writeback is enabled on an inode
253 * @inode: inode of interest
255 * cgroup writeback requires support from both the bdi and filesystem.
256 * Also, both memcg and iocg have to be on the default hierarchy. Test
257 * whether all conditions are met.
259 * Note that the test result may change dynamically on the same inode
260 * depending on how memcg and iocg are configured.
262 static inline bool inode_cgwb_enabled(struct inode *inode)
264 struct backing_dev_info *bdi = inode_to_bdi(inode);
266 return cgroup_on_dfl(mem_cgroup_root_css->cgroup) &&
267 cgroup_on_dfl(blkcg_root_css->cgroup) &&
268 bdi_cap_account_dirty(bdi) &&
269 (bdi->capabilities & BDI_CAP_CGROUP_WRITEBACK) &&
270 (inode->i_sb->s_iflags & SB_I_CGROUPWB);
274 * wb_find_current - find wb for %current on a bdi
275 * @bdi: bdi of interest
277 * Find the wb of @bdi which matches both the memcg and blkcg of %current.
278 * Must be called under rcu_read_lock() which protects the returend wb.
281 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
283 struct cgroup_subsys_state *memcg_css;
284 struct bdi_writeback *wb;
286 memcg_css = task_css(current, memory_cgrp_id);
287 if (!memcg_css->parent)
290 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
293 * %current's blkcg equals the effective blkcg of its memcg. No
294 * need to use the relatively expensive cgroup_get_e_css().
296 if (likely(wb && wb->blkcg_css == task_css(current, io_cgrp_id)))
302 * wb_get_create_current - get or create wb for %current on a bdi
303 * @bdi: bdi of interest
304 * @gfp: allocation mask
306 * Equivalent to wb_get_create() on %current's memcg. This function is
307 * called from a relatively hot path and optimizes the common cases using
310 static inline struct bdi_writeback *
311 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
313 struct bdi_writeback *wb;
316 wb = wb_find_current(bdi);
317 if (wb && unlikely(!wb_tryget(wb)))
322 struct cgroup_subsys_state *memcg_css;
324 memcg_css = task_get_css(current, memory_cgrp_id);
325 wb = wb_get_create(bdi, memcg_css, gfp);
332 * inode_to_wb_is_valid - test whether an inode has a wb associated
333 * @inode: inode of interest
335 * Returns %true if @inode has a wb associated. May be called without any
338 static inline bool inode_to_wb_is_valid(struct inode *inode)
344 * inode_to_wb - determine the wb of an inode
345 * @inode: inode of interest
347 * Returns the wb @inode is currently associated with. The caller must be
348 * holding either @inode->i_lock, @inode->i_mapping->tree_lock, or the
349 * associated wb's list_lock.
351 static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
353 #ifdef CONFIG_LOCKDEP
354 WARN_ON_ONCE(debug_locks &&
355 (!lockdep_is_held(&inode->i_lock) &&
356 !lockdep_is_held(&inode->i_mapping->tree_lock) &&
357 !lockdep_is_held(&inode->i_wb->list_lock)));
363 * unlocked_inode_to_wb_begin - begin unlocked inode wb access transaction
364 * @inode: target inode
365 * @lockedp: temp bool output param, to be passed to the end function
367 * The caller wants to access the wb associated with @inode but isn't
368 * holding inode->i_lock, mapping->tree_lock or wb->list_lock. This
369 * function determines the wb associated with @inode and ensures that the
370 * association doesn't change until the transaction is finished with
371 * unlocked_inode_to_wb_end().
373 * The caller must call unlocked_inode_to_wb_end() with *@lockdep
374 * afterwards and can't sleep during transaction. IRQ may or may not be
375 * disabled on return.
377 static inline struct bdi_writeback *
378 unlocked_inode_to_wb_begin(struct inode *inode, bool *lockedp)
383 * Paired with store_release in inode_switch_wb_work_fn() and
384 * ensures that we see the new wb if we see cleared I_WB_SWITCH.
386 *lockedp = smp_load_acquire(&inode->i_state) & I_WB_SWITCH;
388 if (unlikely(*lockedp))
389 spin_lock_irq(&inode->i_mapping->tree_lock);
392 * Protected by either !I_WB_SWITCH + rcu_read_lock() or tree_lock.
393 * inode_to_wb() will bark. Deref directly.
399 * unlocked_inode_to_wb_end - end inode wb access transaction
400 * @inode: target inode
401 * @locked: *@lockedp from unlocked_inode_to_wb_begin()
403 static inline void unlocked_inode_to_wb_end(struct inode *inode, bool locked)
405 if (unlikely(locked))
406 spin_unlock_irq(&inode->i_mapping->tree_lock);
413 struct radix_tree_iter tree_iter;
417 static inline struct bdi_writeback *__wb_iter_next(struct wb_iter *iter,
418 struct backing_dev_info *bdi)
420 struct radix_tree_iter *titer = &iter->tree_iter;
422 WARN_ON_ONCE(!rcu_read_lock_held());
424 if (iter->start_memcg_id >= 0) {
425 iter->slot = radix_tree_iter_init(titer, iter->start_memcg_id);
426 iter->start_memcg_id = -1;
428 iter->slot = radix_tree_next_slot(iter->slot, titer, 0);
432 iter->slot = radix_tree_next_chunk(&bdi->cgwb_tree, titer, 0);
438 static inline struct bdi_writeback *__wb_iter_init(struct wb_iter *iter,
439 struct backing_dev_info *bdi,
442 iter->start_memcg_id = start_memcg_id;
445 return __wb_iter_next(iter, bdi);
451 * bdi_for_each_wb - walk all wb's of a bdi in ascending memcg ID order
452 * @wb_cur: cursor struct bdi_writeback pointer
453 * @bdi: bdi to walk wb's of
454 * @iter: pointer to struct wb_iter to be used as iteration buffer
455 * @start_memcg_id: memcg ID to start iteration from
457 * Iterate @wb_cur through the wb's (bdi_writeback's) of @bdi in ascending
458 * memcg ID order starting from @start_memcg_id. @iter is struct wb_iter
459 * to be used as temp storage during iteration. rcu_read_lock() must be
460 * held throughout iteration.
462 #define bdi_for_each_wb(wb_cur, bdi, iter, start_memcg_id) \
463 for ((wb_cur) = __wb_iter_init(iter, bdi, start_memcg_id); \
464 (wb_cur); (wb_cur) = __wb_iter_next(iter, bdi))
466 #else /* CONFIG_CGROUP_WRITEBACK */
468 static inline bool inode_cgwb_enabled(struct inode *inode)
473 static inline struct bdi_writeback_congested *
474 wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp)
476 atomic_inc(&bdi->wb_congested->refcnt);
477 return bdi->wb_congested;
480 static inline void wb_congested_put(struct bdi_writeback_congested *congested)
482 if (atomic_dec_and_test(&congested->refcnt))
486 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
491 static inline struct bdi_writeback *
492 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
497 static inline bool inode_to_wb_is_valid(struct inode *inode)
502 static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
504 return &inode_to_bdi(inode)->wb;
507 static inline struct bdi_writeback *
508 unlocked_inode_to_wb_begin(struct inode *inode, bool *lockedp)
510 return inode_to_wb(inode);
513 static inline void unlocked_inode_to_wb_end(struct inode *inode, bool locked)
517 static inline void wb_memcg_offline(struct mem_cgroup *memcg)
521 static inline void wb_blkcg_offline(struct blkcg *blkcg)
529 #define bdi_for_each_wb(wb_cur, bdi, iter, start_blkcg_id) \
530 for ((iter)->next_id = (start_blkcg_id); \
531 ({ (wb_cur) = !(iter)->next_id++ ? &(bdi)->wb : NULL; }); )
533 static inline int inode_congested(struct inode *inode, int cong_bits)
535 return wb_congested(&inode_to_bdi(inode)->wb, cong_bits);
538 #endif /* CONFIG_CGROUP_WRITEBACK */
540 static inline int inode_read_congested(struct inode *inode)
542 return inode_congested(inode, 1 << WB_sync_congested);
545 static inline int inode_write_congested(struct inode *inode)
547 return inode_congested(inode, 1 << WB_async_congested);
550 static inline int inode_rw_congested(struct inode *inode)
552 return inode_congested(inode, (1 << WB_sync_congested) |
553 (1 << WB_async_congested));
556 static inline int bdi_congested(struct backing_dev_info *bdi, int cong_bits)
558 return wb_congested(&bdi->wb, cong_bits);
561 static inline int bdi_read_congested(struct backing_dev_info *bdi)
563 return bdi_congested(bdi, 1 << WB_sync_congested);
566 static inline int bdi_write_congested(struct backing_dev_info *bdi)
568 return bdi_congested(bdi, 1 << WB_async_congested);
571 static inline int bdi_rw_congested(struct backing_dev_info *bdi)
573 return bdi_congested(bdi, (1 << WB_sync_congested) |
574 (1 << WB_async_congested));
577 #endif /* _LINUX_BACKING_DEV_H */