1 // SPDX-License-Identifier: GPL-2.0-only
3 #include <linux/blkdev.h>
4 #include <linux/wait.h>
5 #include <linux/rbtree.h>
6 #include <linux/kthread.h>
7 #include <linux/backing-dev.h>
8 #include <linux/blk-cgroup.h>
9 #include <linux/freezer.h>
11 #include <linux/pagemap.h>
13 #include <linux/sched/mm.h>
14 #include <linux/sched.h>
15 #include <linux/module.h>
16 #include <linux/writeback.h>
17 #include <linux/device.h>
18 #include <trace/events/writeback.h>
21 struct backing_dev_info noop_backing_dev_info
;
22 EXPORT_SYMBOL_GPL(noop_backing_dev_info
);
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
);
76 "BdiWriteback: %10lu kB\n"
77 "BdiReclaimable: %10lu kB\n"
78 "BdiDirtyThresh: %10lu kB\n"
79 "DirtyThresh: %10lu kB\n"
80 "BackgroundThresh: %10lu kB\n"
81 "BdiDirtied: %10lu kB\n"
82 "BdiWritten: %10lu kB\n"
83 "BdiWriteBandwidth: %10lu kBps\n"
87 "b_dirty_time: %10lu\n"
90 (unsigned long) K(wb_stat(wb
, WB_WRITEBACK
)),
91 (unsigned long) K(wb_stat(wb
, WB_RECLAIMABLE
)),
95 (unsigned long) K(wb_stat(wb
, WB_DIRTIED
)),
96 (unsigned long) K(wb_stat(wb
, WB_WRITTEN
)),
97 (unsigned long) K(wb
->write_bandwidth
),
102 !list_empty(&bdi
->bdi_list
), bdi
->wb
.state
);
106 DEFINE_SHOW_ATTRIBUTE(bdi_debug_stats
);
108 static void bdi_debug_register(struct backing_dev_info
*bdi
, const char *name
)
110 bdi
->debug_dir
= debugfs_create_dir(name
, bdi_debug_root
);
112 debugfs_create_file("stats", 0444, bdi
->debug_dir
, bdi
,
113 &bdi_debug_stats_fops
);
116 static void bdi_debug_unregister(struct backing_dev_info
*bdi
)
118 debugfs_remove_recursive(bdi
->debug_dir
);
121 static inline void bdi_debug_init(void)
124 static inline void bdi_debug_register(struct backing_dev_info
*bdi
,
128 static inline void bdi_debug_unregister(struct backing_dev_info
*bdi
)
133 static ssize_t
read_ahead_kb_store(struct device
*dev
,
134 struct device_attribute
*attr
,
135 const char *buf
, size_t count
)
137 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
138 unsigned long read_ahead_kb
;
141 ret
= kstrtoul(buf
, 10, &read_ahead_kb
);
145 bdi
->ra_pages
= read_ahead_kb
>> (PAGE_SHIFT
- 10);
150 #define BDI_SHOW(name, expr) \
151 static ssize_t name##_show(struct device *dev, \
152 struct device_attribute *attr, char *buf) \
154 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
156 return sysfs_emit(buf, "%lld\n", (long long)expr); \
158 static DEVICE_ATTR_RW(name);
160 BDI_SHOW(read_ahead_kb
, K(bdi
->ra_pages
))
162 static ssize_t
min_ratio_store(struct device
*dev
,
163 struct device_attribute
*attr
, const char *buf
, size_t count
)
165 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
169 ret
= kstrtouint(buf
, 10, &ratio
);
173 ret
= bdi_set_min_ratio(bdi
, ratio
);
179 BDI_SHOW(min_ratio
, bdi
->min_ratio
/ BDI_RATIO_SCALE
)
181 static ssize_t
min_ratio_fine_store(struct device
*dev
,
182 struct device_attribute
*attr
, const char *buf
, size_t count
)
184 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
188 ret
= kstrtouint(buf
, 10, &ratio
);
192 ret
= bdi_set_min_ratio_no_scale(bdi
, ratio
);
198 BDI_SHOW(min_ratio_fine
, bdi
->min_ratio
)
200 static ssize_t
max_ratio_store(struct device
*dev
,
201 struct device_attribute
*attr
, const char *buf
, size_t count
)
203 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
207 ret
= kstrtouint(buf
, 10, &ratio
);
211 ret
= bdi_set_max_ratio(bdi
, ratio
);
217 BDI_SHOW(max_ratio
, bdi
->max_ratio
/ BDI_RATIO_SCALE
)
219 static ssize_t
max_ratio_fine_store(struct device
*dev
,
220 struct device_attribute
*attr
, const char *buf
, size_t count
)
222 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
226 ret
= kstrtouint(buf
, 10, &ratio
);
230 ret
= bdi_set_max_ratio_no_scale(bdi
, ratio
);
236 BDI_SHOW(max_ratio_fine
, bdi
->max_ratio
)
238 static ssize_t
min_bytes_show(struct device
*dev
,
239 struct device_attribute
*attr
,
242 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
244 return sysfs_emit(buf
, "%llu\n", bdi_get_min_bytes(bdi
));
247 static ssize_t
min_bytes_store(struct device
*dev
,
248 struct device_attribute
*attr
, const char *buf
, size_t count
)
250 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
254 ret
= kstrtoull(buf
, 10, &bytes
);
258 ret
= bdi_set_min_bytes(bdi
, bytes
);
264 static DEVICE_ATTR_RW(min_bytes
);
266 static ssize_t
max_bytes_show(struct device
*dev
,
267 struct device_attribute
*attr
,
270 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
272 return sysfs_emit(buf
, "%llu\n", bdi_get_max_bytes(bdi
));
275 static ssize_t
max_bytes_store(struct device
*dev
,
276 struct device_attribute
*attr
, const char *buf
, size_t count
)
278 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
282 ret
= kstrtoull(buf
, 10, &bytes
);
286 ret
= bdi_set_max_bytes(bdi
, bytes
);
292 static DEVICE_ATTR_RW(max_bytes
);
294 static ssize_t
stable_pages_required_show(struct device
*dev
,
295 struct device_attribute
*attr
,
299 "the stable_pages_required attribute has been removed. Use the stable_writes queue attribute instead.\n");
300 return sysfs_emit(buf
, "%d\n", 0);
302 static DEVICE_ATTR_RO(stable_pages_required
);
304 static ssize_t
strict_limit_store(struct device
*dev
,
305 struct device_attribute
*attr
, const char *buf
, size_t count
)
307 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
308 unsigned int strict_limit
;
311 ret
= kstrtouint(buf
, 10, &strict_limit
);
315 ret
= bdi_set_strict_limit(bdi
, strict_limit
);
322 static ssize_t
strict_limit_show(struct device
*dev
,
323 struct device_attribute
*attr
, char *buf
)
325 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
327 return sysfs_emit(buf
, "%d\n",
328 !!(bdi
->capabilities
& BDI_CAP_STRICTLIMIT
));
330 static DEVICE_ATTR_RW(strict_limit
);
332 static struct attribute
*bdi_dev_attrs
[] = {
333 &dev_attr_read_ahead_kb
.attr
,
334 &dev_attr_min_ratio
.attr
,
335 &dev_attr_min_ratio_fine
.attr
,
336 &dev_attr_max_ratio
.attr
,
337 &dev_attr_max_ratio_fine
.attr
,
338 &dev_attr_min_bytes
.attr
,
339 &dev_attr_max_bytes
.attr
,
340 &dev_attr_stable_pages_required
.attr
,
341 &dev_attr_strict_limit
.attr
,
344 ATTRIBUTE_GROUPS(bdi_dev
);
346 static const struct class bdi_class
= {
348 .dev_groups
= bdi_dev_groups
,
351 static __init
int bdi_class_init(void)
355 ret
= class_register(&bdi_class
);
363 postcore_initcall(bdi_class_init
);
365 static int __init
default_bdi_init(void)
367 bdi_wq
= alloc_workqueue("writeback", WQ_MEM_RECLAIM
| WQ_UNBOUND
|
373 subsys_initcall(default_bdi_init
);
376 * This function is used when the first inode for this wb is marked dirty. It
377 * wakes-up the corresponding bdi thread which should then take care of the
378 * periodic background write-out of dirty inodes. Since the write-out would
379 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
380 * set up a timer which wakes the bdi thread up later.
382 * Note, we wouldn't bother setting up the timer, but this function is on the
383 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
384 * by delaying the wake-up.
386 * We have to be careful not to postpone flush work if it is scheduled for
387 * earlier. Thus we use queue_delayed_work().
389 void wb_wakeup_delayed(struct bdi_writeback
*wb
)
391 unsigned long timeout
;
393 timeout
= msecs_to_jiffies(dirty_writeback_interval
* 10);
394 spin_lock_irq(&wb
->work_lock
);
395 if (test_bit(WB_registered
, &wb
->state
))
396 queue_delayed_work(bdi_wq
, &wb
->dwork
, timeout
);
397 spin_unlock_irq(&wb
->work_lock
);
400 static void wb_update_bandwidth_workfn(struct work_struct
*work
)
402 struct bdi_writeback
*wb
= container_of(to_delayed_work(work
),
403 struct bdi_writeback
, bw_dwork
);
405 wb_update_bandwidth(wb
);
409 * Initial write bandwidth: 100 MB/s
411 #define INIT_BW (100 << (20 - PAGE_SHIFT))
413 static int wb_init(struct bdi_writeback
*wb
, struct backing_dev_info
*bdi
,
418 memset(wb
, 0, sizeof(*wb
));
421 wb
->last_old_flush
= jiffies
;
422 INIT_LIST_HEAD(&wb
->b_dirty
);
423 INIT_LIST_HEAD(&wb
->b_io
);
424 INIT_LIST_HEAD(&wb
->b_more_io
);
425 INIT_LIST_HEAD(&wb
->b_dirty_time
);
426 spin_lock_init(&wb
->list_lock
);
428 atomic_set(&wb
->writeback_inodes
, 0);
429 wb
->bw_time_stamp
= jiffies
;
430 wb
->balanced_dirty_ratelimit
= INIT_BW
;
431 wb
->dirty_ratelimit
= INIT_BW
;
432 wb
->write_bandwidth
= INIT_BW
;
433 wb
->avg_write_bandwidth
= INIT_BW
;
435 spin_lock_init(&wb
->work_lock
);
436 INIT_LIST_HEAD(&wb
->work_list
);
437 INIT_DELAYED_WORK(&wb
->dwork
, wb_workfn
);
438 INIT_DELAYED_WORK(&wb
->bw_dwork
, wb_update_bandwidth_workfn
);
439 wb
->dirty_sleep
= jiffies
;
441 err
= fprop_local_init_percpu(&wb
->completions
, gfp
);
445 for (i
= 0; i
< NR_WB_STAT_ITEMS
; i
++) {
446 err
= percpu_counter_init(&wb
->stat
[i
], 0, gfp
);
448 goto out_destroy_stat
;
455 percpu_counter_destroy(&wb
->stat
[i
]);
456 fprop_local_destroy_percpu(&wb
->completions
);
460 static void cgwb_remove_from_bdi_list(struct bdi_writeback
*wb
);
463 * Remove bdi from the global list and shutdown any threads we have running
465 static void wb_shutdown(struct bdi_writeback
*wb
)
467 /* Make sure nobody queues further work */
468 spin_lock_irq(&wb
->work_lock
);
469 if (!test_and_clear_bit(WB_registered
, &wb
->state
)) {
470 spin_unlock_irq(&wb
->work_lock
);
473 spin_unlock_irq(&wb
->work_lock
);
475 cgwb_remove_from_bdi_list(wb
);
477 * Drain work list and shutdown the delayed_work. !WB_registered
478 * tells wb_workfn() that @wb is dying and its work_list needs to
479 * be drained no matter what.
481 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
482 flush_delayed_work(&wb
->dwork
);
483 WARN_ON(!list_empty(&wb
->work_list
));
484 flush_delayed_work(&wb
->bw_dwork
);
487 static void wb_exit(struct bdi_writeback
*wb
)
491 WARN_ON(delayed_work_pending(&wb
->dwork
));
493 for (i
= 0; i
< NR_WB_STAT_ITEMS
; i
++)
494 percpu_counter_destroy(&wb
->stat
[i
]);
496 fprop_local_destroy_percpu(&wb
->completions
);
499 #ifdef CONFIG_CGROUP_WRITEBACK
501 #include <linux/memcontrol.h>
504 * cgwb_lock protects bdi->cgwb_tree, blkcg->cgwb_list, offline_cgwbs and
505 * memcg->cgwb_list. bdi->cgwb_tree is also RCU protected.
507 static DEFINE_SPINLOCK(cgwb_lock
);
508 static struct workqueue_struct
*cgwb_release_wq
;
510 static LIST_HEAD(offline_cgwbs
);
511 static void cleanup_offline_cgwbs_workfn(struct work_struct
*work
);
512 static DECLARE_WORK(cleanup_offline_cgwbs_work
, cleanup_offline_cgwbs_workfn
);
514 static void cgwb_free_rcu(struct rcu_head
*rcu_head
)
516 struct bdi_writeback
*wb
= container_of(rcu_head
,
517 struct bdi_writeback
, rcu
);
519 percpu_ref_exit(&wb
->refcnt
);
523 static void cgwb_release_workfn(struct work_struct
*work
)
525 struct bdi_writeback
*wb
= container_of(work
, struct bdi_writeback
,
527 struct backing_dev_info
*bdi
= wb
->bdi
;
529 mutex_lock(&wb
->bdi
->cgwb_release_mutex
);
532 css_put(wb
->memcg_css
);
533 css_put(wb
->blkcg_css
);
534 mutex_unlock(&wb
->bdi
->cgwb_release_mutex
);
536 /* triggers blkg destruction if no online users left */
537 blkcg_unpin_online(wb
->blkcg_css
);
539 fprop_local_destroy_percpu(&wb
->memcg_completions
);
541 spin_lock_irq(&cgwb_lock
);
542 list_del(&wb
->offline_node
);
543 spin_unlock_irq(&cgwb_lock
);
547 WARN_ON_ONCE(!list_empty(&wb
->b_attached
));
548 call_rcu(&wb
->rcu
, cgwb_free_rcu
);
551 static void cgwb_release(struct percpu_ref
*refcnt
)
553 struct bdi_writeback
*wb
= container_of(refcnt
, struct bdi_writeback
,
555 queue_work(cgwb_release_wq
, &wb
->release_work
);
558 static void cgwb_kill(struct bdi_writeback
*wb
)
560 lockdep_assert_held(&cgwb_lock
);
562 WARN_ON(!radix_tree_delete(&wb
->bdi
->cgwb_tree
, wb
->memcg_css
->id
));
563 list_del(&wb
->memcg_node
);
564 list_del(&wb
->blkcg_node
);
565 list_add(&wb
->offline_node
, &offline_cgwbs
);
566 percpu_ref_kill(&wb
->refcnt
);
569 static void cgwb_remove_from_bdi_list(struct bdi_writeback
*wb
)
571 spin_lock_irq(&cgwb_lock
);
572 list_del_rcu(&wb
->bdi_node
);
573 spin_unlock_irq(&cgwb_lock
);
576 static int cgwb_create(struct backing_dev_info
*bdi
,
577 struct cgroup_subsys_state
*memcg_css
, gfp_t gfp
)
579 struct mem_cgroup
*memcg
;
580 struct cgroup_subsys_state
*blkcg_css
;
581 struct list_head
*memcg_cgwb_list
, *blkcg_cgwb_list
;
582 struct bdi_writeback
*wb
;
586 memcg
= mem_cgroup_from_css(memcg_css
);
587 blkcg_css
= cgroup_get_e_css(memcg_css
->cgroup
, &io_cgrp_subsys
);
588 memcg_cgwb_list
= &memcg
->cgwb_list
;
589 blkcg_cgwb_list
= blkcg_get_cgwb_list(blkcg_css
);
591 /* look up again under lock and discard on blkcg mismatch */
592 spin_lock_irqsave(&cgwb_lock
, flags
);
593 wb
= radix_tree_lookup(&bdi
->cgwb_tree
, memcg_css
->id
);
594 if (wb
&& wb
->blkcg_css
!= blkcg_css
) {
598 spin_unlock_irqrestore(&cgwb_lock
, flags
);
602 /* need to create a new one */
603 wb
= kmalloc(sizeof(*wb
), gfp
);
609 ret
= wb_init(wb
, bdi
, gfp
);
613 ret
= percpu_ref_init(&wb
->refcnt
, cgwb_release
, 0, gfp
);
617 ret
= fprop_local_init_percpu(&wb
->memcg_completions
, gfp
);
621 wb
->memcg_css
= memcg_css
;
622 wb
->blkcg_css
= blkcg_css
;
623 INIT_LIST_HEAD(&wb
->b_attached
);
624 INIT_WORK(&wb
->release_work
, cgwb_release_workfn
);
625 set_bit(WB_registered
, &wb
->state
);
629 * The root wb determines the registered state of the whole bdi and
630 * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate
631 * whether they're still online. Don't link @wb if any is dead.
632 * See wb_memcg_offline() and wb_blkcg_offline().
635 spin_lock_irqsave(&cgwb_lock
, flags
);
636 if (test_bit(WB_registered
, &bdi
->wb
.state
) &&
637 blkcg_cgwb_list
->next
&& memcg_cgwb_list
->next
) {
638 /* we might have raced another instance of this function */
639 ret
= radix_tree_insert(&bdi
->cgwb_tree
, memcg_css
->id
, wb
);
641 list_add_tail_rcu(&wb
->bdi_node
, &bdi
->wb_list
);
642 list_add(&wb
->memcg_node
, memcg_cgwb_list
);
643 list_add(&wb
->blkcg_node
, blkcg_cgwb_list
);
644 blkcg_pin_online(blkcg_css
);
649 spin_unlock_irqrestore(&cgwb_lock
, flags
);
659 fprop_local_destroy_percpu(&wb
->memcg_completions
);
661 percpu_ref_exit(&wb
->refcnt
);
672 * wb_get_lookup - get wb for a given memcg
674 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
676 * Try to get the wb for @memcg_css on @bdi. The returned wb has its
677 * refcount incremented.
679 * This function uses css_get() on @memcg_css and thus expects its refcnt
680 * to be positive on invocation. IOW, rcu_read_lock() protection on
681 * @memcg_css isn't enough. try_get it before calling this function.
683 * A wb is keyed by its associated memcg. As blkcg implicitly enables
684 * memcg on the default hierarchy, memcg association is guaranteed to be
685 * more specific (equal or descendant to the associated blkcg) and thus can
686 * identify both the memcg and blkcg associations.
688 * Because the blkcg associated with a memcg may change as blkcg is enabled
689 * and disabled closer to root in the hierarchy, each wb keeps track of
690 * both the memcg and blkcg associated with it and verifies the blkcg on
691 * each lookup. On mismatch, the existing wb is discarded and a new one is
694 struct bdi_writeback
*wb_get_lookup(struct backing_dev_info
*bdi
,
695 struct cgroup_subsys_state
*memcg_css
)
697 struct bdi_writeback
*wb
;
699 if (!memcg_css
->parent
)
703 wb
= radix_tree_lookup(&bdi
->cgwb_tree
, memcg_css
->id
);
705 struct cgroup_subsys_state
*blkcg_css
;
707 /* see whether the blkcg association has changed */
708 blkcg_css
= cgroup_get_e_css(memcg_css
->cgroup
, &io_cgrp_subsys
);
709 if (unlikely(wb
->blkcg_css
!= blkcg_css
|| !wb_tryget(wb
)))
719 * wb_get_create - get wb for a given memcg, create if necessary
721 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
722 * @gfp: allocation mask to use
724 * Try to get the wb for @memcg_css on @bdi. If it doesn't exist, try to
725 * create one. See wb_get_lookup() for more details.
727 struct bdi_writeback
*wb_get_create(struct backing_dev_info
*bdi
,
728 struct cgroup_subsys_state
*memcg_css
,
731 struct bdi_writeback
*wb
;
736 wb
= wb_get_lookup(bdi
, memcg_css
);
737 } while (!wb
&& !cgwb_create(bdi
, memcg_css
, gfp
));
742 static int cgwb_bdi_init(struct backing_dev_info
*bdi
)
746 INIT_RADIX_TREE(&bdi
->cgwb_tree
, GFP_ATOMIC
);
747 mutex_init(&bdi
->cgwb_release_mutex
);
748 init_rwsem(&bdi
->wb_switch_rwsem
);
750 ret
= wb_init(&bdi
->wb
, bdi
, GFP_KERNEL
);
752 bdi
->wb
.memcg_css
= &root_mem_cgroup
->css
;
753 bdi
->wb
.blkcg_css
= blkcg_root_css
;
758 static void cgwb_bdi_unregister(struct backing_dev_info
*bdi
)
760 struct radix_tree_iter iter
;
762 struct bdi_writeback
*wb
;
764 WARN_ON(test_bit(WB_registered
, &bdi
->wb
.state
));
766 spin_lock_irq(&cgwb_lock
);
767 radix_tree_for_each_slot(slot
, &bdi
->cgwb_tree
, &iter
, 0)
769 spin_unlock_irq(&cgwb_lock
);
771 mutex_lock(&bdi
->cgwb_release_mutex
);
772 spin_lock_irq(&cgwb_lock
);
773 while (!list_empty(&bdi
->wb_list
)) {
774 wb
= list_first_entry(&bdi
->wb_list
, struct bdi_writeback
,
776 spin_unlock_irq(&cgwb_lock
);
778 spin_lock_irq(&cgwb_lock
);
780 spin_unlock_irq(&cgwb_lock
);
781 mutex_unlock(&bdi
->cgwb_release_mutex
);
785 * cleanup_offline_cgwbs_workfn - try to release dying cgwbs
787 * Try to release dying cgwbs by switching attached inodes to the nearest
788 * living ancestor's writeback. Processed wbs are placed at the end
789 * of the list to guarantee the forward progress.
791 static void cleanup_offline_cgwbs_workfn(struct work_struct
*work
)
793 struct bdi_writeback
*wb
;
794 LIST_HEAD(processed
);
796 spin_lock_irq(&cgwb_lock
);
798 while (!list_empty(&offline_cgwbs
)) {
799 wb
= list_first_entry(&offline_cgwbs
, struct bdi_writeback
,
801 list_move(&wb
->offline_node
, &processed
);
804 * If wb is dirty, cleaning up the writeback by switching
805 * attached inodes will result in an effective removal of any
806 * bandwidth restrictions, which isn't the goal. Instead,
807 * it can be postponed until the next time, when all io
808 * will be likely completed. If in the meantime some inodes
809 * will get re-dirtied, they should be eventually switched to
812 if (wb_has_dirty_io(wb
))
818 spin_unlock_irq(&cgwb_lock
);
819 while (cleanup_offline_cgwb(wb
))
821 spin_lock_irq(&cgwb_lock
);
826 if (!list_empty(&processed
))
827 list_splice_tail(&processed
, &offline_cgwbs
);
829 spin_unlock_irq(&cgwb_lock
);
833 * wb_memcg_offline - kill all wb's associated with a memcg being offlined
834 * @memcg: memcg being offlined
836 * Also prevents creation of any new wb's associated with @memcg.
838 void wb_memcg_offline(struct mem_cgroup
*memcg
)
840 struct list_head
*memcg_cgwb_list
= &memcg
->cgwb_list
;
841 struct bdi_writeback
*wb
, *next
;
843 spin_lock_irq(&cgwb_lock
);
844 list_for_each_entry_safe(wb
, next
, memcg_cgwb_list
, memcg_node
)
846 memcg_cgwb_list
->next
= NULL
; /* prevent new wb's */
847 spin_unlock_irq(&cgwb_lock
);
849 queue_work(system_unbound_wq
, &cleanup_offline_cgwbs_work
);
853 * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined
854 * @css: blkcg being offlined
856 * Also prevents creation of any new wb's associated with @blkcg.
858 void wb_blkcg_offline(struct cgroup_subsys_state
*css
)
860 struct bdi_writeback
*wb
, *next
;
861 struct list_head
*list
= blkcg_get_cgwb_list(css
);
863 spin_lock_irq(&cgwb_lock
);
864 list_for_each_entry_safe(wb
, next
, list
, blkcg_node
)
866 list
->next
= NULL
; /* prevent new wb's */
867 spin_unlock_irq(&cgwb_lock
);
870 static void cgwb_bdi_register(struct backing_dev_info
*bdi
)
872 spin_lock_irq(&cgwb_lock
);
873 list_add_tail_rcu(&bdi
->wb
.bdi_node
, &bdi
->wb_list
);
874 spin_unlock_irq(&cgwb_lock
);
877 static int __init
cgwb_init(void)
880 * There can be many concurrent release work items overwhelming
881 * system_wq. Put them in a separate wq and limit concurrency.
882 * There's no point in executing many of these in parallel.
884 cgwb_release_wq
= alloc_workqueue("cgwb_release", 0, 1);
885 if (!cgwb_release_wq
)
890 subsys_initcall(cgwb_init
);
892 #else /* CONFIG_CGROUP_WRITEBACK */
894 static int cgwb_bdi_init(struct backing_dev_info
*bdi
)
896 return wb_init(&bdi
->wb
, bdi
, GFP_KERNEL
);
899 static void cgwb_bdi_unregister(struct backing_dev_info
*bdi
) { }
901 static void cgwb_bdi_register(struct backing_dev_info
*bdi
)
903 list_add_tail_rcu(&bdi
->wb
.bdi_node
, &bdi
->wb_list
);
906 static void cgwb_remove_from_bdi_list(struct bdi_writeback
*wb
)
908 list_del_rcu(&wb
->bdi_node
);
911 #endif /* CONFIG_CGROUP_WRITEBACK */
913 int bdi_init(struct backing_dev_info
*bdi
)
917 kref_init(&bdi
->refcnt
);
919 bdi
->max_ratio
= 100 * BDI_RATIO_SCALE
;
920 bdi
->max_prop_frac
= FPROP_FRAC_BASE
;
921 INIT_LIST_HEAD(&bdi
->bdi_list
);
922 INIT_LIST_HEAD(&bdi
->wb_list
);
923 init_waitqueue_head(&bdi
->wb_waitq
);
925 return cgwb_bdi_init(bdi
);
928 struct backing_dev_info
*bdi_alloc(int node_id
)
930 struct backing_dev_info
*bdi
;
932 bdi
= kzalloc_node(sizeof(*bdi
), GFP_KERNEL
, node_id
);
940 bdi
->capabilities
= BDI_CAP_WRITEBACK
| BDI_CAP_WRITEBACK_ACCT
;
941 bdi
->ra_pages
= VM_READAHEAD_PAGES
;
942 bdi
->io_pages
= VM_READAHEAD_PAGES
;
943 timer_setup(&bdi
->laptop_mode_wb_timer
, laptop_mode_timer_fn
, 0);
946 EXPORT_SYMBOL(bdi_alloc
);
948 static struct rb_node
**bdi_lookup_rb_node(u64 id
, struct rb_node
**parentp
)
950 struct rb_node
**p
= &bdi_tree
.rb_node
;
951 struct rb_node
*parent
= NULL
;
952 struct backing_dev_info
*bdi
;
954 lockdep_assert_held(&bdi_lock
);
958 bdi
= rb_entry(parent
, struct backing_dev_info
, rb_node
);
962 else if (bdi
->id
< id
)
974 * bdi_get_by_id - lookup and get bdi from its id
975 * @id: bdi id to lookup
977 * Find bdi matching @id and get it. Returns NULL if the matching bdi
978 * doesn't exist or is already unregistered.
980 struct backing_dev_info
*bdi_get_by_id(u64 id
)
982 struct backing_dev_info
*bdi
= NULL
;
985 spin_lock_bh(&bdi_lock
);
986 p
= bdi_lookup_rb_node(id
, NULL
);
988 bdi
= rb_entry(*p
, struct backing_dev_info
, rb_node
);
991 spin_unlock_bh(&bdi_lock
);
996 int bdi_register_va(struct backing_dev_info
*bdi
, const char *fmt
, va_list args
)
999 struct rb_node
*parent
, **p
;
1001 if (bdi
->dev
) /* The driver needs to use separate queues per device */
1004 vsnprintf(bdi
->dev_name
, sizeof(bdi
->dev_name
), fmt
, args
);
1005 dev
= device_create(&bdi_class
, NULL
, MKDEV(0, 0), bdi
, bdi
->dev_name
);
1007 return PTR_ERR(dev
);
1009 cgwb_bdi_register(bdi
);
1012 bdi_debug_register(bdi
, dev_name(dev
));
1013 set_bit(WB_registered
, &bdi
->wb
.state
);
1015 spin_lock_bh(&bdi_lock
);
1017 bdi
->id
= ++bdi_id_cursor
;
1019 p
= bdi_lookup_rb_node(bdi
->id
, &parent
);
1020 rb_link_node(&bdi
->rb_node
, parent
, p
);
1021 rb_insert_color(&bdi
->rb_node
, &bdi_tree
);
1023 list_add_tail_rcu(&bdi
->bdi_list
, &bdi_list
);
1025 spin_unlock_bh(&bdi_lock
);
1027 trace_writeback_bdi_register(bdi
);
1031 int bdi_register(struct backing_dev_info
*bdi
, const char *fmt
, ...)
1036 va_start(args
, fmt
);
1037 ret
= bdi_register_va(bdi
, fmt
, args
);
1041 EXPORT_SYMBOL(bdi_register
);
1043 void bdi_set_owner(struct backing_dev_info
*bdi
, struct device
*owner
)
1045 WARN_ON_ONCE(bdi
->owner
);
1051 * Remove bdi from bdi_list, and ensure that it is no longer visible
1053 static void bdi_remove_from_list(struct backing_dev_info
*bdi
)
1055 spin_lock_bh(&bdi_lock
);
1056 rb_erase(&bdi
->rb_node
, &bdi_tree
);
1057 list_del_rcu(&bdi
->bdi_list
);
1058 spin_unlock_bh(&bdi_lock
);
1060 synchronize_rcu_expedited();
1063 void bdi_unregister(struct backing_dev_info
*bdi
)
1065 del_timer_sync(&bdi
->laptop_mode_wb_timer
);
1067 /* make sure nobody finds us on the bdi_list anymore */
1068 bdi_remove_from_list(bdi
);
1069 wb_shutdown(&bdi
->wb
);
1070 cgwb_bdi_unregister(bdi
);
1073 * If this BDI's min ratio has been set, use bdi_set_min_ratio() to
1074 * update the global bdi_min_ratio.
1077 bdi_set_min_ratio(bdi
, 0);
1080 bdi_debug_unregister(bdi
);
1081 device_unregister(bdi
->dev
);
1086 put_device(bdi
->owner
);
1090 EXPORT_SYMBOL(bdi_unregister
);
1092 static void release_bdi(struct kref
*ref
)
1094 struct backing_dev_info
*bdi
=
1095 container_of(ref
, struct backing_dev_info
, refcnt
);
1097 WARN_ON_ONCE(test_bit(WB_registered
, &bdi
->wb
.state
));
1098 WARN_ON_ONCE(bdi
->dev
);
1103 void bdi_put(struct backing_dev_info
*bdi
)
1105 kref_put(&bdi
->refcnt
, release_bdi
);
1107 EXPORT_SYMBOL(bdi_put
);
1109 struct backing_dev_info
*inode_to_bdi(struct inode
*inode
)
1111 struct super_block
*sb
;
1114 return &noop_backing_dev_info
;
1118 if (sb_is_blkdev_sb(sb
))
1119 return I_BDEV(inode
)->bd_disk
->bdi
;
1123 EXPORT_SYMBOL(inode_to_bdi
);
1125 const char *bdi_dev_name(struct backing_dev_info
*bdi
)
1127 if (!bdi
|| !bdi
->dev
)
1128 return bdi_unknown_name
;
1129 return bdi
->dev_name
;
1131 EXPORT_SYMBOL_GPL(bdi_dev_name
);