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/blk-cgroup.h>
17 #include <linux/backing-dev-defs.h>
18 #include <linux/slab.h>
20 int __must_check bdi_init(struct backing_dev_info *bdi);
21 void bdi_exit(struct backing_dev_info *bdi);
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);
29 int __must_check bdi_setup_and_register(struct backing_dev_info *, char *);
30 void bdi_destroy(struct backing_dev_info *bdi);
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);
38 extern spinlock_t bdi_lock;
39 extern struct list_head bdi_list;
41 extern struct workqueue_struct *bdi_wq;
43 static inline bool wb_has_dirty_io(struct bdi_writeback *wb)
45 return test_bit(WB_has_dirty_io, &wb->state);
48 static inline bool bdi_has_dirty_io(struct backing_dev_info *bdi)
51 * @bdi->tot_write_bandwidth is guaranteed to be > 0 if there are
52 * any dirty wbs. See wb_update_write_bandwidth().
54 return atomic_long_read(&bdi->tot_write_bandwidth);
57 static inline void __add_wb_stat(struct bdi_writeback *wb,
58 enum wb_stat_item item, s64 amount)
60 __percpu_counter_add(&wb->stat[item], amount, WB_STAT_BATCH);
63 static inline void __inc_wb_stat(struct bdi_writeback *wb,
64 enum wb_stat_item item)
66 __add_wb_stat(wb, item, 1);
69 static inline void inc_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
73 local_irq_save(flags);
74 __inc_wb_stat(wb, item);
75 local_irq_restore(flags);
78 static inline void __dec_wb_stat(struct bdi_writeback *wb,
79 enum wb_stat_item item)
81 __add_wb_stat(wb, item, -1);
84 static inline void dec_wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
88 local_irq_save(flags);
89 __dec_wb_stat(wb, item);
90 local_irq_restore(flags);
93 static inline s64 wb_stat(struct bdi_writeback *wb, enum wb_stat_item item)
95 return percpu_counter_read_positive(&wb->stat[item]);
98 static inline s64 __wb_stat_sum(struct bdi_writeback *wb,
99 enum wb_stat_item item)
101 return percpu_counter_sum_positive(&wb->stat[item]);
104 static inline s64 wb_stat_sum(struct bdi_writeback *wb, enum wb_stat_item item)
109 local_irq_save(flags);
110 sum = __wb_stat_sum(wb, item);
111 local_irq_restore(flags);
116 extern void wb_writeout_inc(struct bdi_writeback *wb);
119 * maximal error of a stat counter.
121 static inline unsigned long wb_stat_error(struct bdi_writeback *wb)
124 return nr_cpu_ids * WB_STAT_BATCH;
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);
134 * Flags in backing_dev_info::capability
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)
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.
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.
149 * BDI_CAP_CGROUP_WRITEBACK: Supports cgroup-aware writeback.
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
158 #define BDI_CAP_NO_ACCT_AND_WRITEBACK \
159 (BDI_CAP_NO_WRITEBACK | BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_ACCT_WB)
161 extern struct backing_dev_info noop_backing_dev_info;
164 * writeback_in_progress - determine whether there is writeback in progress
165 * @wb: bdi_writeback of interest
167 * Determine whether there is writeback waiting to be handled against a
170 static inline bool writeback_in_progress(struct bdi_writeback *wb)
172 return test_bit(WB_writeback_running, &wb->state);
175 static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
177 struct super_block *sb;
180 return &noop_backing_dev_info;
184 if (sb_is_blkdev_sb(sb))
185 return blk_get_backing_dev_info(I_BDEV(inode));
190 static inline int wb_congested(struct bdi_writeback *wb, int cong_bits)
192 struct backing_dev_info *bdi = wb->bdi;
194 if (bdi->congested_fn)
195 return bdi->congested_fn(bdi->congested_data, cong_bits);
196 return wb->congested->state & cong_bits;
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);
204 static inline bool bdi_cap_stable_pages_required(struct backing_dev_info *bdi)
206 return bdi->capabilities & BDI_CAP_STABLE_WRITES;
209 static inline bool bdi_cap_writeback_dirty(struct backing_dev_info *bdi)
211 return !(bdi->capabilities & BDI_CAP_NO_WRITEBACK);
214 static inline bool bdi_cap_account_dirty(struct backing_dev_info *bdi)
216 return !(bdi->capabilities & BDI_CAP_NO_ACCT_DIRTY);
219 static inline bool bdi_cap_account_writeback(struct backing_dev_info *bdi)
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));
226 static inline bool mapping_cap_writeback_dirty(struct address_space *mapping)
228 return bdi_cap_writeback_dirty(inode_to_bdi(mapping->host));
231 static inline bool mapping_cap_account_dirty(struct address_space *mapping)
233 return bdi_cap_account_dirty(inode_to_bdi(mapping->host));
236 static inline int bdi_sched_wait(void *word)
242 #ifdef CONFIG_CGROUP_WRITEBACK
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,
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);
255 * inode_cgwb_enabled - test whether cgroup writeback is enabled on an inode
256 * @inode: inode of interest
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.
262 * Note that the test result may change dynamically on the same inode
263 * depending on how memcg and iocg are configured.
265 static inline bool inode_cgwb_enabled(struct inode *inode)
267 struct backing_dev_info *bdi = inode_to_bdi(inode);
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);
277 * wb_find_current - find wb for %current on a bdi
278 * @bdi: bdi of interest
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.
284 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
286 struct cgroup_subsys_state *memcg_css;
287 struct bdi_writeback *wb;
289 memcg_css = task_css(current, memory_cgrp_id);
290 if (!memcg_css->parent)
293 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
296 * %current's blkcg equals the effective blkcg of its memcg. No
297 * need to use the relatively expensive cgroup_get_e_css().
299 if (likely(wb && wb->blkcg_css == task_css(current, io_cgrp_id)))
305 * wb_get_create_current - get or create wb for %current on a bdi
306 * @bdi: bdi of interest
307 * @gfp: allocation mask
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
313 static inline struct bdi_writeback *
314 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
316 struct bdi_writeback *wb;
319 wb = wb_find_current(bdi);
320 if (wb && unlikely(!wb_tryget(wb)))
325 struct cgroup_subsys_state *memcg_css;
327 memcg_css = task_get_css(current, memory_cgrp_id);
328 wb = wb_get_create(bdi, memcg_css, gfp);
335 * inode_to_wb_is_valid - test whether an inode has a wb associated
336 * @inode: inode of interest
338 * Returns %true if @inode has a wb associated. May be called without any
341 static inline bool inode_to_wb_is_valid(struct inode *inode)
347 * inode_to_wb - determine the wb of an inode
348 * @inode: inode of interest
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.
354 static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
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)));
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
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().
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.
380 static inline struct bdi_writeback *
381 unlocked_inode_to_wb_begin(struct inode *inode, bool *lockedp)
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.
389 *lockedp = smp_load_acquire(&inode->i_state) & I_WB_SWITCH;
391 if (unlikely(*lockedp))
392 spin_lock_irq(&inode->i_mapping->tree_lock);
395 * Protected by either !I_WB_SWITCH + rcu_read_lock() or tree_lock.
396 * inode_to_wb() will bark. Deref directly.
402 * unlocked_inode_to_wb_end - end inode wb access transaction
403 * @inode: target inode
404 * @locked: *@lockedp from unlocked_inode_to_wb_begin()
406 static inline void unlocked_inode_to_wb_end(struct inode *inode, bool locked)
408 if (unlikely(locked))
409 spin_unlock_irq(&inode->i_mapping->tree_lock);
414 #else /* CONFIG_CGROUP_WRITEBACK */
416 static inline bool inode_cgwb_enabled(struct inode *inode)
421 static inline struct bdi_writeback_congested *
422 wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp)
424 atomic_inc(&bdi->wb_congested->refcnt);
425 return bdi->wb_congested;
428 static inline void wb_congested_put(struct bdi_writeback_congested *congested)
430 if (atomic_dec_and_test(&congested->refcnt))
434 static inline struct bdi_writeback *wb_find_current(struct backing_dev_info *bdi)
439 static inline struct bdi_writeback *
440 wb_get_create_current(struct backing_dev_info *bdi, gfp_t gfp)
445 static inline bool inode_to_wb_is_valid(struct inode *inode)
450 static inline struct bdi_writeback *inode_to_wb(struct inode *inode)
452 return &inode_to_bdi(inode)->wb;
455 static inline struct bdi_writeback *
456 unlocked_inode_to_wb_begin(struct inode *inode, bool *lockedp)
458 return inode_to_wb(inode);
461 static inline void unlocked_inode_to_wb_end(struct inode *inode, bool locked)
465 static inline void wb_memcg_offline(struct mem_cgroup *memcg)
469 static inline void wb_blkcg_offline(struct blkcg *blkcg)
473 static inline int inode_congested(struct inode *inode, int cong_bits)
475 return wb_congested(&inode_to_bdi(inode)->wb, cong_bits);
478 #endif /* CONFIG_CGROUP_WRITEBACK */
480 static inline int inode_read_congested(struct inode *inode)
482 return inode_congested(inode, 1 << WB_sync_congested);
485 static inline int inode_write_congested(struct inode *inode)
487 return inode_congested(inode, 1 << WB_async_congested);
490 static inline int inode_rw_congested(struct inode *inode)
492 return inode_congested(inode, (1 << WB_sync_congested) |
493 (1 << WB_async_congested));
496 static inline int bdi_congested(struct backing_dev_info *bdi, int cong_bits)
498 return wb_congested(&bdi->wb, cong_bits);
501 static inline int bdi_read_congested(struct backing_dev_info *bdi)
503 return bdi_congested(bdi, 1 << WB_sync_congested);
506 static inline int bdi_write_congested(struct backing_dev_info *bdi)
508 return bdi_congested(bdi, 1 << WB_async_congested);
511 static inline int bdi_rw_congested(struct backing_dev_info *bdi)
513 return bdi_congested(bdi, (1 << WB_sync_congested) |
514 (1 << WB_async_congested));
517 #endif /* _LINUX_BACKING_DEV_H */