1 // SPDX-License-Identifier: GPL-2.0-only
3 #include <linux/wait.h>
4 #include <linux/rbtree.h>
5 #include <linux/backing-dev.h>
6 #include <linux/kthread.h>
7 #include <linux/freezer.h>
9 #include <linux/pagemap.h>
11 #include <linux/sched.h>
12 #include <linux/module.h>
13 #include <linux/writeback.h>
14 #include <linux/device.h>
15 #include <trace/events/writeback.h>
17 struct backing_dev_info noop_backing_dev_info
= {
19 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
21 EXPORT_SYMBOL_GPL(noop_backing_dev_info
);
23 static struct class *bdi_class
;
24 static const char *bdi_unknown_name
= "(unknown)";
27 * bdi_lock protects bdi_tree and updates to bdi_list. bdi_list has RCU
28 * reader side locking.
30 DEFINE_SPINLOCK(bdi_lock
);
31 static u64 bdi_id_cursor
;
32 static struct rb_root bdi_tree
= RB_ROOT
;
35 /* bdi_wq serves all asynchronous writeback tasks */
36 struct workqueue_struct
*bdi_wq
;
38 #ifdef CONFIG_DEBUG_FS
39 #include <linux/debugfs.h>
40 #include <linux/seq_file.h>
42 static struct dentry
*bdi_debug_root
;
44 static void bdi_debug_init(void)
46 bdi_debug_root
= debugfs_create_dir("bdi", NULL
);
49 static int bdi_debug_stats_show(struct seq_file
*m
, void *v
)
51 struct backing_dev_info
*bdi
= m
->private;
52 struct bdi_writeback
*wb
= &bdi
->wb
;
53 unsigned long background_thresh
;
54 unsigned long dirty_thresh
;
55 unsigned long wb_thresh
;
56 unsigned long nr_dirty
, nr_io
, nr_more_io
, nr_dirty_time
;
59 nr_dirty
= nr_io
= nr_more_io
= nr_dirty_time
= 0;
60 spin_lock(&wb
->list_lock
);
61 list_for_each_entry(inode
, &wb
->b_dirty
, i_io_list
)
63 list_for_each_entry(inode
, &wb
->b_io
, i_io_list
)
65 list_for_each_entry(inode
, &wb
->b_more_io
, i_io_list
)
67 list_for_each_entry(inode
, &wb
->b_dirty_time
, i_io_list
)
68 if (inode
->i_state
& I_DIRTY_TIME
)
70 spin_unlock(&wb
->list_lock
);
72 global_dirty_limits(&background_thresh
, &dirty_thresh
);
73 wb_thresh
= wb_calc_thresh(wb
, dirty_thresh
);
75 #define K(x) ((x) << (PAGE_SHIFT - 10))
77 "BdiWriteback: %10lu kB\n"
78 "BdiReclaimable: %10lu kB\n"
79 "BdiDirtyThresh: %10lu kB\n"
80 "DirtyThresh: %10lu kB\n"
81 "BackgroundThresh: %10lu kB\n"
82 "BdiDirtied: %10lu kB\n"
83 "BdiWritten: %10lu kB\n"
84 "BdiWriteBandwidth: %10lu kBps\n"
88 "b_dirty_time: %10lu\n"
91 (unsigned long) K(wb_stat(wb
, WB_WRITEBACK
)),
92 (unsigned long) K(wb_stat(wb
, WB_RECLAIMABLE
)),
96 (unsigned long) K(wb_stat(wb
, WB_DIRTIED
)),
97 (unsigned long) K(wb_stat(wb
, WB_WRITTEN
)),
98 (unsigned long) K(wb
->write_bandwidth
),
103 !list_empty(&bdi
->bdi_list
), bdi
->wb
.state
);
108 DEFINE_SHOW_ATTRIBUTE(bdi_debug_stats
);
110 static void bdi_debug_register(struct backing_dev_info
*bdi
, const char *name
)
112 bdi
->debug_dir
= debugfs_create_dir(name
, bdi_debug_root
);
114 debugfs_create_file("stats", 0444, bdi
->debug_dir
, bdi
,
115 &bdi_debug_stats_fops
);
118 static void bdi_debug_unregister(struct backing_dev_info
*bdi
)
120 debugfs_remove_recursive(bdi
->debug_dir
);
123 static inline void bdi_debug_init(void)
126 static inline void bdi_debug_register(struct backing_dev_info
*bdi
,
130 static inline void bdi_debug_unregister(struct backing_dev_info
*bdi
)
135 static ssize_t
read_ahead_kb_store(struct device
*dev
,
136 struct device_attribute
*attr
,
137 const char *buf
, size_t count
)
139 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
140 unsigned long read_ahead_kb
;
143 ret
= kstrtoul(buf
, 10, &read_ahead_kb
);
147 bdi
->ra_pages
= read_ahead_kb
>> (PAGE_SHIFT
- 10);
152 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
154 #define BDI_SHOW(name, expr) \
155 static ssize_t name##_show(struct device *dev, \
156 struct device_attribute *attr, char *page) \
158 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
160 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
162 static DEVICE_ATTR_RW(name);
164 BDI_SHOW(read_ahead_kb
, K(bdi
->ra_pages
))
166 static ssize_t
min_ratio_store(struct device
*dev
,
167 struct device_attribute
*attr
, const char *buf
, size_t count
)
169 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
173 ret
= kstrtouint(buf
, 10, &ratio
);
177 ret
= bdi_set_min_ratio(bdi
, ratio
);
183 BDI_SHOW(min_ratio
, bdi
->min_ratio
)
185 static ssize_t
max_ratio_store(struct device
*dev
,
186 struct device_attribute
*attr
, const char *buf
, size_t count
)
188 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
192 ret
= kstrtouint(buf
, 10, &ratio
);
196 ret
= bdi_set_max_ratio(bdi
, ratio
);
202 BDI_SHOW(max_ratio
, bdi
->max_ratio
)
204 static ssize_t
stable_pages_required_show(struct device
*dev
,
205 struct device_attribute
*attr
,
208 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
210 return snprintf(page
, PAGE_SIZE
-1, "%d\n",
211 bdi_cap_stable_pages_required(bdi
) ? 1 : 0);
213 static DEVICE_ATTR_RO(stable_pages_required
);
215 static struct attribute
*bdi_dev_attrs
[] = {
216 &dev_attr_read_ahead_kb
.attr
,
217 &dev_attr_min_ratio
.attr
,
218 &dev_attr_max_ratio
.attr
,
219 &dev_attr_stable_pages_required
.attr
,
222 ATTRIBUTE_GROUPS(bdi_dev
);
224 static __init
int bdi_class_init(void)
226 bdi_class
= class_create(THIS_MODULE
, "bdi");
227 if (IS_ERR(bdi_class
))
228 return PTR_ERR(bdi_class
);
230 bdi_class
->dev_groups
= bdi_dev_groups
;
235 postcore_initcall(bdi_class_init
);
237 static int bdi_init(struct backing_dev_info
*bdi
);
239 static int __init
default_bdi_init(void)
243 bdi_wq
= alloc_workqueue("writeback", WQ_MEM_RECLAIM
| WQ_UNBOUND
|
248 err
= bdi_init(&noop_backing_dev_info
);
252 subsys_initcall(default_bdi_init
);
255 * This function is used when the first inode for this wb is marked dirty. It
256 * wakes-up the corresponding bdi thread which should then take care of the
257 * periodic background write-out of dirty inodes. Since the write-out would
258 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
259 * set up a timer which wakes the bdi thread up later.
261 * Note, we wouldn't bother setting up the timer, but this function is on the
262 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
263 * by delaying the wake-up.
265 * We have to be careful not to postpone flush work if it is scheduled for
266 * earlier. Thus we use queue_delayed_work().
268 void wb_wakeup_delayed(struct bdi_writeback
*wb
)
270 unsigned long timeout
;
272 timeout
= msecs_to_jiffies(dirty_writeback_interval
* 10);
273 spin_lock_bh(&wb
->work_lock
);
274 if (test_bit(WB_registered
, &wb
->state
))
275 queue_delayed_work(bdi_wq
, &wb
->dwork
, timeout
);
276 spin_unlock_bh(&wb
->work_lock
);
280 * Initial write bandwidth: 100 MB/s
282 #define INIT_BW (100 << (20 - PAGE_SHIFT))
284 static int wb_init(struct bdi_writeback
*wb
, struct backing_dev_info
*bdi
,
285 int blkcg_id
, gfp_t gfp
)
289 memset(wb
, 0, sizeof(*wb
));
294 wb
->last_old_flush
= jiffies
;
295 INIT_LIST_HEAD(&wb
->b_dirty
);
296 INIT_LIST_HEAD(&wb
->b_io
);
297 INIT_LIST_HEAD(&wb
->b_more_io
);
298 INIT_LIST_HEAD(&wb
->b_dirty_time
);
299 spin_lock_init(&wb
->list_lock
);
301 wb
->bw_time_stamp
= jiffies
;
302 wb
->balanced_dirty_ratelimit
= INIT_BW
;
303 wb
->dirty_ratelimit
= INIT_BW
;
304 wb
->write_bandwidth
= INIT_BW
;
305 wb
->avg_write_bandwidth
= INIT_BW
;
307 spin_lock_init(&wb
->work_lock
);
308 INIT_LIST_HEAD(&wb
->work_list
);
309 INIT_DELAYED_WORK(&wb
->dwork
, wb_workfn
);
310 wb
->dirty_sleep
= jiffies
;
312 wb
->congested
= wb_congested_get_create(bdi
, blkcg_id
, gfp
);
313 if (!wb
->congested
) {
318 err
= fprop_local_init_percpu(&wb
->completions
, gfp
);
322 for (i
= 0; i
< NR_WB_STAT_ITEMS
; i
++) {
323 err
= percpu_counter_init(&wb
->stat
[i
], 0, gfp
);
325 goto out_destroy_stat
;
332 percpu_counter_destroy(&wb
->stat
[i
]);
333 fprop_local_destroy_percpu(&wb
->completions
);
335 wb_congested_put(wb
->congested
);
342 static void cgwb_remove_from_bdi_list(struct bdi_writeback
*wb
);
345 * Remove bdi from the global list and shutdown any threads we have running
347 static void wb_shutdown(struct bdi_writeback
*wb
)
349 /* Make sure nobody queues further work */
350 spin_lock_bh(&wb
->work_lock
);
351 if (!test_and_clear_bit(WB_registered
, &wb
->state
)) {
352 spin_unlock_bh(&wb
->work_lock
);
355 spin_unlock_bh(&wb
->work_lock
);
357 cgwb_remove_from_bdi_list(wb
);
359 * Drain work list and shutdown the delayed_work. !WB_registered
360 * tells wb_workfn() that @wb is dying and its work_list needs to
361 * be drained no matter what.
363 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
364 flush_delayed_work(&wb
->dwork
);
365 WARN_ON(!list_empty(&wb
->work_list
));
368 static void wb_exit(struct bdi_writeback
*wb
)
372 WARN_ON(delayed_work_pending(&wb
->dwork
));
374 for (i
= 0; i
< NR_WB_STAT_ITEMS
; i
++)
375 percpu_counter_destroy(&wb
->stat
[i
]);
377 fprop_local_destroy_percpu(&wb
->completions
);
378 wb_congested_put(wb
->congested
);
379 if (wb
!= &wb
->bdi
->wb
)
383 #ifdef CONFIG_CGROUP_WRITEBACK
385 #include <linux/memcontrol.h>
388 * cgwb_lock protects bdi->cgwb_tree, bdi->cgwb_congested_tree,
389 * blkcg->cgwb_list, and memcg->cgwb_list. bdi->cgwb_tree is also RCU
392 static DEFINE_SPINLOCK(cgwb_lock
);
393 static struct workqueue_struct
*cgwb_release_wq
;
396 * wb_congested_get_create - get or create a wb_congested
397 * @bdi: associated bdi
398 * @blkcg_id: ID of the associated blkcg
399 * @gfp: allocation mask
401 * Look up the wb_congested for @blkcg_id on @bdi. If missing, create one.
402 * The returned wb_congested has its reference count incremented. Returns
405 struct bdi_writeback_congested
*
406 wb_congested_get_create(struct backing_dev_info
*bdi
, int blkcg_id
, gfp_t gfp
)
408 struct bdi_writeback_congested
*new_congested
= NULL
, *congested
;
409 struct rb_node
**node
, *parent
;
412 spin_lock_irqsave(&cgwb_lock
, flags
);
414 node
= &bdi
->cgwb_congested_tree
.rb_node
;
417 while (*node
!= NULL
) {
419 congested
= rb_entry(parent
, struct bdi_writeback_congested
,
421 if (congested
->blkcg_id
< blkcg_id
)
422 node
= &parent
->rb_left
;
423 else if (congested
->blkcg_id
> blkcg_id
)
424 node
= &parent
->rb_right
;
430 /* !found and storage for new one already allocated, insert */
431 congested
= new_congested
;
432 rb_link_node(&congested
->rb_node
, parent
, node
);
433 rb_insert_color(&congested
->rb_node
, &bdi
->cgwb_congested_tree
);
434 spin_unlock_irqrestore(&cgwb_lock
, flags
);
438 spin_unlock_irqrestore(&cgwb_lock
, flags
);
440 /* allocate storage for new one and retry */
441 new_congested
= kzalloc(sizeof(*new_congested
), gfp
);
445 refcount_set(&new_congested
->refcnt
, 1);
446 new_congested
->__bdi
= bdi
;
447 new_congested
->blkcg_id
= blkcg_id
;
451 refcount_inc(&congested
->refcnt
);
452 spin_unlock_irqrestore(&cgwb_lock
, flags
);
453 kfree(new_congested
);
458 * wb_congested_put - put a wb_congested
459 * @congested: wb_congested to put
461 * Put @congested and destroy it if the refcnt reaches zero.
463 void wb_congested_put(struct bdi_writeback_congested
*congested
)
467 if (!refcount_dec_and_lock_irqsave(&congested
->refcnt
, &cgwb_lock
, &flags
))
470 /* bdi might already have been destroyed leaving @congested unlinked */
471 if (congested
->__bdi
) {
472 rb_erase(&congested
->rb_node
,
473 &congested
->__bdi
->cgwb_congested_tree
);
474 congested
->__bdi
= NULL
;
477 spin_unlock_irqrestore(&cgwb_lock
, flags
);
481 static void cgwb_release_workfn(struct work_struct
*work
)
483 struct bdi_writeback
*wb
= container_of(work
, struct bdi_writeback
,
485 struct blkcg
*blkcg
= css_to_blkcg(wb
->blkcg_css
);
487 mutex_lock(&wb
->bdi
->cgwb_release_mutex
);
490 css_put(wb
->memcg_css
);
491 css_put(wb
->blkcg_css
);
492 mutex_unlock(&wb
->bdi
->cgwb_release_mutex
);
494 /* triggers blkg destruction if no online users left */
495 blkcg_unpin_online(blkcg
);
497 fprop_local_destroy_percpu(&wb
->memcg_completions
);
498 percpu_ref_exit(&wb
->refcnt
);
503 static void cgwb_release(struct percpu_ref
*refcnt
)
505 struct bdi_writeback
*wb
= container_of(refcnt
, struct bdi_writeback
,
507 queue_work(cgwb_release_wq
, &wb
->release_work
);
510 static void cgwb_kill(struct bdi_writeback
*wb
)
512 lockdep_assert_held(&cgwb_lock
);
514 WARN_ON(!radix_tree_delete(&wb
->bdi
->cgwb_tree
, wb
->memcg_css
->id
));
515 list_del(&wb
->memcg_node
);
516 list_del(&wb
->blkcg_node
);
517 percpu_ref_kill(&wb
->refcnt
);
520 static void cgwb_remove_from_bdi_list(struct bdi_writeback
*wb
)
522 spin_lock_irq(&cgwb_lock
);
523 list_del_rcu(&wb
->bdi_node
);
524 spin_unlock_irq(&cgwb_lock
);
527 static int cgwb_create(struct backing_dev_info
*bdi
,
528 struct cgroup_subsys_state
*memcg_css
, gfp_t gfp
)
530 struct mem_cgroup
*memcg
;
531 struct cgroup_subsys_state
*blkcg_css
;
533 struct list_head
*memcg_cgwb_list
, *blkcg_cgwb_list
;
534 struct bdi_writeback
*wb
;
538 memcg
= mem_cgroup_from_css(memcg_css
);
539 blkcg_css
= cgroup_get_e_css(memcg_css
->cgroup
, &io_cgrp_subsys
);
540 blkcg
= css_to_blkcg(blkcg_css
);
541 memcg_cgwb_list
= &memcg
->cgwb_list
;
542 blkcg_cgwb_list
= &blkcg
->cgwb_list
;
544 /* look up again under lock and discard on blkcg mismatch */
545 spin_lock_irqsave(&cgwb_lock
, flags
);
546 wb
= radix_tree_lookup(&bdi
->cgwb_tree
, memcg_css
->id
);
547 if (wb
&& wb
->blkcg_css
!= blkcg_css
) {
551 spin_unlock_irqrestore(&cgwb_lock
, flags
);
555 /* need to create a new one */
556 wb
= kmalloc(sizeof(*wb
), gfp
);
562 ret
= wb_init(wb
, bdi
, blkcg_css
->id
, gfp
);
566 ret
= percpu_ref_init(&wb
->refcnt
, cgwb_release
, 0, gfp
);
570 ret
= fprop_local_init_percpu(&wb
->memcg_completions
, gfp
);
574 wb
->memcg_css
= memcg_css
;
575 wb
->blkcg_css
= blkcg_css
;
576 INIT_WORK(&wb
->release_work
, cgwb_release_workfn
);
577 set_bit(WB_registered
, &wb
->state
);
580 * The root wb determines the registered state of the whole bdi and
581 * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate
582 * whether they're still online. Don't link @wb if any is dead.
583 * See wb_memcg_offline() and wb_blkcg_offline().
586 spin_lock_irqsave(&cgwb_lock
, flags
);
587 if (test_bit(WB_registered
, &bdi
->wb
.state
) &&
588 blkcg_cgwb_list
->next
&& memcg_cgwb_list
->next
) {
589 /* we might have raced another instance of this function */
590 ret
= radix_tree_insert(&bdi
->cgwb_tree
, memcg_css
->id
, wb
);
592 list_add_tail_rcu(&wb
->bdi_node
, &bdi
->wb_list
);
593 list_add(&wb
->memcg_node
, memcg_cgwb_list
);
594 list_add(&wb
->blkcg_node
, blkcg_cgwb_list
);
595 blkcg_pin_online(blkcg
);
600 spin_unlock_irqrestore(&cgwb_lock
, flags
);
609 fprop_local_destroy_percpu(&wb
->memcg_completions
);
611 percpu_ref_exit(&wb
->refcnt
);
622 * wb_get_lookup - get wb for a given memcg
624 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
626 * Try to get the wb for @memcg_css on @bdi. The returned wb has its
627 * refcount incremented.
629 * This function uses css_get() on @memcg_css and thus expects its refcnt
630 * to be positive on invocation. IOW, rcu_read_lock() protection on
631 * @memcg_css isn't enough. try_get it before calling this function.
633 * A wb is keyed by its associated memcg. As blkcg implicitly enables
634 * memcg on the default hierarchy, memcg association is guaranteed to be
635 * more specific (equal or descendant to the associated blkcg) and thus can
636 * identify both the memcg and blkcg associations.
638 * Because the blkcg associated with a memcg may change as blkcg is enabled
639 * and disabled closer to root in the hierarchy, each wb keeps track of
640 * both the memcg and blkcg associated with it and verifies the blkcg on
641 * each lookup. On mismatch, the existing wb is discarded and a new one is
644 struct bdi_writeback
*wb_get_lookup(struct backing_dev_info
*bdi
,
645 struct cgroup_subsys_state
*memcg_css
)
647 struct bdi_writeback
*wb
;
649 if (!memcg_css
->parent
)
653 wb
= radix_tree_lookup(&bdi
->cgwb_tree
, memcg_css
->id
);
655 struct cgroup_subsys_state
*blkcg_css
;
657 /* see whether the blkcg association has changed */
658 blkcg_css
= cgroup_get_e_css(memcg_css
->cgroup
, &io_cgrp_subsys
);
659 if (unlikely(wb
->blkcg_css
!= blkcg_css
|| !wb_tryget(wb
)))
669 * wb_get_create - get wb for a given memcg, create if necessary
671 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
672 * @gfp: allocation mask to use
674 * Try to get the wb for @memcg_css on @bdi. If it doesn't exist, try to
675 * create one. See wb_get_lookup() for more details.
677 struct bdi_writeback
*wb_get_create(struct backing_dev_info
*bdi
,
678 struct cgroup_subsys_state
*memcg_css
,
681 struct bdi_writeback
*wb
;
683 might_sleep_if(gfpflags_allow_blocking(gfp
));
685 if (!memcg_css
->parent
)
689 wb
= wb_get_lookup(bdi
, memcg_css
);
690 } while (!wb
&& !cgwb_create(bdi
, memcg_css
, gfp
));
695 static int cgwb_bdi_init(struct backing_dev_info
*bdi
)
699 INIT_RADIX_TREE(&bdi
->cgwb_tree
, GFP_ATOMIC
);
700 bdi
->cgwb_congested_tree
= RB_ROOT
;
701 mutex_init(&bdi
->cgwb_release_mutex
);
702 init_rwsem(&bdi
->wb_switch_rwsem
);
704 ret
= wb_init(&bdi
->wb
, bdi
, 1, GFP_KERNEL
);
706 bdi
->wb
.memcg_css
= &root_mem_cgroup
->css
;
707 bdi
->wb
.blkcg_css
= blkcg_root_css
;
712 static void cgwb_bdi_unregister(struct backing_dev_info
*bdi
)
714 struct radix_tree_iter iter
;
716 struct bdi_writeback
*wb
;
718 WARN_ON(test_bit(WB_registered
, &bdi
->wb
.state
));
720 spin_lock_irq(&cgwb_lock
);
721 radix_tree_for_each_slot(slot
, &bdi
->cgwb_tree
, &iter
, 0)
723 spin_unlock_irq(&cgwb_lock
);
725 mutex_lock(&bdi
->cgwb_release_mutex
);
726 spin_lock_irq(&cgwb_lock
);
727 while (!list_empty(&bdi
->wb_list
)) {
728 wb
= list_first_entry(&bdi
->wb_list
, struct bdi_writeback
,
730 spin_unlock_irq(&cgwb_lock
);
732 spin_lock_irq(&cgwb_lock
);
734 spin_unlock_irq(&cgwb_lock
);
735 mutex_unlock(&bdi
->cgwb_release_mutex
);
739 * wb_memcg_offline - kill all wb's associated with a memcg being offlined
740 * @memcg: memcg being offlined
742 * Also prevents creation of any new wb's associated with @memcg.
744 void wb_memcg_offline(struct mem_cgroup
*memcg
)
746 struct list_head
*memcg_cgwb_list
= &memcg
->cgwb_list
;
747 struct bdi_writeback
*wb
, *next
;
749 spin_lock_irq(&cgwb_lock
);
750 list_for_each_entry_safe(wb
, next
, memcg_cgwb_list
, memcg_node
)
752 memcg_cgwb_list
->next
= NULL
; /* prevent new wb's */
753 spin_unlock_irq(&cgwb_lock
);
757 * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined
758 * @blkcg: blkcg being offlined
760 * Also prevents creation of any new wb's associated with @blkcg.
762 void wb_blkcg_offline(struct blkcg
*blkcg
)
764 struct bdi_writeback
*wb
, *next
;
766 spin_lock_irq(&cgwb_lock
);
767 list_for_each_entry_safe(wb
, next
, &blkcg
->cgwb_list
, blkcg_node
)
769 blkcg
->cgwb_list
.next
= NULL
; /* prevent new wb's */
770 spin_unlock_irq(&cgwb_lock
);
773 static void cgwb_bdi_exit(struct backing_dev_info
*bdi
)
777 spin_lock_irq(&cgwb_lock
);
778 while ((rbn
= rb_first(&bdi
->cgwb_congested_tree
))) {
779 struct bdi_writeback_congested
*congested
=
780 rb_entry(rbn
, struct bdi_writeback_congested
, rb_node
);
782 rb_erase(rbn
, &bdi
->cgwb_congested_tree
);
783 congested
->__bdi
= NULL
; /* mark @congested unlinked */
785 spin_unlock_irq(&cgwb_lock
);
788 static void cgwb_bdi_register(struct backing_dev_info
*bdi
)
790 spin_lock_irq(&cgwb_lock
);
791 list_add_tail_rcu(&bdi
->wb
.bdi_node
, &bdi
->wb_list
);
792 spin_unlock_irq(&cgwb_lock
);
795 static int __init
cgwb_init(void)
798 * There can be many concurrent release work items overwhelming
799 * system_wq. Put them in a separate wq and limit concurrency.
800 * There's no point in executing many of these in parallel.
802 cgwb_release_wq
= alloc_workqueue("cgwb_release", 0, 1);
803 if (!cgwb_release_wq
)
808 subsys_initcall(cgwb_init
);
810 #else /* CONFIG_CGROUP_WRITEBACK */
812 static int cgwb_bdi_init(struct backing_dev_info
*bdi
)
816 bdi
->wb_congested
= kzalloc(sizeof(*bdi
->wb_congested
), GFP_KERNEL
);
817 if (!bdi
->wb_congested
)
820 refcount_set(&bdi
->wb_congested
->refcnt
, 1);
822 err
= wb_init(&bdi
->wb
, bdi
, 1, GFP_KERNEL
);
824 wb_congested_put(bdi
->wb_congested
);
830 static void cgwb_bdi_unregister(struct backing_dev_info
*bdi
) { }
832 static void cgwb_bdi_exit(struct backing_dev_info
*bdi
)
834 wb_congested_put(bdi
->wb_congested
);
837 static void cgwb_bdi_register(struct backing_dev_info
*bdi
)
839 list_add_tail_rcu(&bdi
->wb
.bdi_node
, &bdi
->wb_list
);
842 static void cgwb_remove_from_bdi_list(struct bdi_writeback
*wb
)
844 list_del_rcu(&wb
->bdi_node
);
847 #endif /* CONFIG_CGROUP_WRITEBACK */
849 static int bdi_init(struct backing_dev_info
*bdi
)
855 kref_init(&bdi
->refcnt
);
857 bdi
->max_ratio
= 100;
858 bdi
->max_prop_frac
= FPROP_FRAC_BASE
;
859 INIT_LIST_HEAD(&bdi
->bdi_list
);
860 INIT_LIST_HEAD(&bdi
->wb_list
);
861 init_waitqueue_head(&bdi
->wb_waitq
);
863 ret
= cgwb_bdi_init(bdi
);
868 struct backing_dev_info
*bdi_alloc_node(gfp_t gfp_mask
, int node_id
)
870 struct backing_dev_info
*bdi
;
872 bdi
= kmalloc_node(sizeof(struct backing_dev_info
),
873 gfp_mask
| __GFP_ZERO
, node_id
);
883 EXPORT_SYMBOL(bdi_alloc_node
);
885 static struct rb_node
**bdi_lookup_rb_node(u64 id
, struct rb_node
**parentp
)
887 struct rb_node
**p
= &bdi_tree
.rb_node
;
888 struct rb_node
*parent
= NULL
;
889 struct backing_dev_info
*bdi
;
891 lockdep_assert_held(&bdi_lock
);
895 bdi
= rb_entry(parent
, struct backing_dev_info
, rb_node
);
899 else if (bdi
->id
< id
)
911 * bdi_get_by_id - lookup and get bdi from its id
912 * @id: bdi id to lookup
914 * Find bdi matching @id and get it. Returns NULL if the matching bdi
915 * doesn't exist or is already unregistered.
917 struct backing_dev_info
*bdi_get_by_id(u64 id
)
919 struct backing_dev_info
*bdi
= NULL
;
922 spin_lock_bh(&bdi_lock
);
923 p
= bdi_lookup_rb_node(id
, NULL
);
925 bdi
= rb_entry(*p
, struct backing_dev_info
, rb_node
);
928 spin_unlock_bh(&bdi_lock
);
933 int bdi_register_va(struct backing_dev_info
*bdi
, const char *fmt
, va_list args
)
936 struct rb_node
*parent
, **p
;
938 if (bdi
->dev
) /* The driver needs to use separate queues per device */
941 vsnprintf(bdi
->dev_name
, sizeof(bdi
->dev_name
), fmt
, args
);
942 dev
= device_create(bdi_class
, NULL
, MKDEV(0, 0), bdi
, bdi
->dev_name
);
946 cgwb_bdi_register(bdi
);
949 bdi_debug_register(bdi
, dev_name(dev
));
950 set_bit(WB_registered
, &bdi
->wb
.state
);
952 spin_lock_bh(&bdi_lock
);
954 bdi
->id
= ++bdi_id_cursor
;
956 p
= bdi_lookup_rb_node(bdi
->id
, &parent
);
957 rb_link_node(&bdi
->rb_node
, parent
, p
);
958 rb_insert_color(&bdi
->rb_node
, &bdi_tree
);
960 list_add_tail_rcu(&bdi
->bdi_list
, &bdi_list
);
962 spin_unlock_bh(&bdi_lock
);
964 trace_writeback_bdi_register(bdi
);
967 EXPORT_SYMBOL(bdi_register_va
);
969 int bdi_register(struct backing_dev_info
*bdi
, const char *fmt
, ...)
975 ret
= bdi_register_va(bdi
, fmt
, args
);
979 EXPORT_SYMBOL(bdi_register
);
981 int bdi_register_owner(struct backing_dev_info
*bdi
, struct device
*owner
)
985 rc
= bdi_register(bdi
, "%u:%u", MAJOR(owner
->devt
), MINOR(owner
->devt
));
988 /* Leaking owner reference... */
994 EXPORT_SYMBOL(bdi_register_owner
);
997 * Remove bdi from bdi_list, and ensure that it is no longer visible
999 static void bdi_remove_from_list(struct backing_dev_info
*bdi
)
1001 spin_lock_bh(&bdi_lock
);
1002 rb_erase(&bdi
->rb_node
, &bdi_tree
);
1003 list_del_rcu(&bdi
->bdi_list
);
1004 spin_unlock_bh(&bdi_lock
);
1006 synchronize_rcu_expedited();
1009 void bdi_unregister(struct backing_dev_info
*bdi
)
1011 /* make sure nobody finds us on the bdi_list anymore */
1012 bdi_remove_from_list(bdi
);
1013 wb_shutdown(&bdi
->wb
);
1014 cgwb_bdi_unregister(bdi
);
1017 bdi_debug_unregister(bdi
);
1018 device_unregister(bdi
->dev
);
1023 put_device(bdi
->owner
);
1028 static void release_bdi(struct kref
*ref
)
1030 struct backing_dev_info
*bdi
=
1031 container_of(ref
, struct backing_dev_info
, refcnt
);
1033 if (test_bit(WB_registered
, &bdi
->wb
.state
))
1034 bdi_unregister(bdi
);
1035 WARN_ON_ONCE(bdi
->dev
);
1041 void bdi_put(struct backing_dev_info
*bdi
)
1043 kref_put(&bdi
->refcnt
, release_bdi
);
1045 EXPORT_SYMBOL(bdi_put
);
1047 const char *bdi_dev_name(struct backing_dev_info
*bdi
)
1049 if (!bdi
|| !bdi
->dev
)
1050 return bdi_unknown_name
;
1051 return bdi
->dev_name
;
1053 EXPORT_SYMBOL_GPL(bdi_dev_name
);
1055 static wait_queue_head_t congestion_wqh
[2] = {
1056 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[0]),
1057 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[1])
1059 static atomic_t nr_wb_congested
[2];
1061 void clear_wb_congested(struct bdi_writeback_congested
*congested
, int sync
)
1063 wait_queue_head_t
*wqh
= &congestion_wqh
[sync
];
1064 enum wb_congested_state bit
;
1066 bit
= sync
? WB_sync_congested
: WB_async_congested
;
1067 if (test_and_clear_bit(bit
, &congested
->state
))
1068 atomic_dec(&nr_wb_congested
[sync
]);
1069 smp_mb__after_atomic();
1070 if (waitqueue_active(wqh
))
1073 EXPORT_SYMBOL(clear_wb_congested
);
1075 void set_wb_congested(struct bdi_writeback_congested
*congested
, int sync
)
1077 enum wb_congested_state bit
;
1079 bit
= sync
? WB_sync_congested
: WB_async_congested
;
1080 if (!test_and_set_bit(bit
, &congested
->state
))
1081 atomic_inc(&nr_wb_congested
[sync
]);
1083 EXPORT_SYMBOL(set_wb_congested
);
1086 * congestion_wait - wait for a backing_dev to become uncongested
1087 * @sync: SYNC or ASYNC IO
1088 * @timeout: timeout in jiffies
1090 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
1091 * write congestion. If no backing_devs are congested then just wait for the
1092 * next write to be completed.
1094 long congestion_wait(int sync
, long timeout
)
1097 unsigned long start
= jiffies
;
1099 wait_queue_head_t
*wqh
= &congestion_wqh
[sync
];
1101 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
1102 ret
= io_schedule_timeout(timeout
);
1103 finish_wait(wqh
, &wait
);
1105 trace_writeback_congestion_wait(jiffies_to_usecs(timeout
),
1106 jiffies_to_usecs(jiffies
- start
));
1110 EXPORT_SYMBOL(congestion_wait
);
1113 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a pgdat to complete writes
1114 * @sync: SYNC or ASYNC IO
1115 * @timeout: timeout in jiffies
1117 * In the event of a congested backing_dev (any backing_dev) this waits
1118 * for up to @timeout jiffies for either a BDI to exit congestion of the
1119 * given @sync queue or a write to complete.
1121 * The return value is 0 if the sleep is for the full timeout. Otherwise,
1122 * it is the number of jiffies that were still remaining when the function
1123 * returned. return_value == timeout implies the function did not sleep.
1125 long wait_iff_congested(int sync
, long timeout
)
1128 unsigned long start
= jiffies
;
1130 wait_queue_head_t
*wqh
= &congestion_wqh
[sync
];
1133 * If there is no congestion, yield if necessary instead
1134 * of sleeping on the congestion queue
1136 if (atomic_read(&nr_wb_congested
[sync
]) == 0) {
1139 /* In case we scheduled, work out time remaining */
1140 ret
= timeout
- (jiffies
- start
);
1147 /* Sleep until uncongested or a write happens */
1148 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
1149 ret
= io_schedule_timeout(timeout
);
1150 finish_wait(wqh
, &wait
);
1153 trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout
),
1154 jiffies_to_usecs(jiffies
- start
));
1158 EXPORT_SYMBOL(wait_iff_congested
);