path: root/block/blk-cgroup.c (plain)
blob: f570f387034dd53b446f26be5564f7c8ea6005ec

1	/*
2	* Common Block IO controller cgroup interface
3	*
4	* Based on ideas and code from CFQ, CFS and BFQ:
5	* Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
6	*
7	* Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
8	* Paolo Valente <paolo.valente@unimore.it>
9	*
10	* Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
11	* Nauman Rafique <nauman@google.com>
12	*
13	* For policy-specific per-blkcg data:
14	* Copyright (C) 2015 Paolo Valente <paolo.valente@unimore.it>
15	* Arianna Avanzini <avanzini.arianna@gmail.com>
16	*/
17	#include <linux/ioprio.h>
18	#include <linux/kdev_t.h>
19	#include <linux/module.h>
20	#include <linux/err.h>
21	#include <linux/blkdev.h>
22	#include <linux/backing-dev.h>
23	#include <linux/slab.h>
24	#include <linux/genhd.h>
25	#include <linux/delay.h>
26	#include <linux/atomic.h>
27	#include <linux/ctype.h>
28	#include <linux/blk-cgroup.h>
29	#include "blk.h"
30
31	#define MAX_KEY_LEN 100
32
33	/*
34	* blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation.
35	* blkcg_pol_register_mutex nests outside of it and synchronizes entire
36	* policy [un]register operations including cgroup file additions /
37	* removals. Putting cgroup file registration outside blkcg_pol_mutex
38	* allows grabbing it from cgroup callbacks.
39	*/
40	static DEFINE_MUTEX(blkcg_pol_register_mutex);
41	static DEFINE_MUTEX(blkcg_pol_mutex);
42
43	struct blkcg blkcg_root;
44	EXPORT_SYMBOL_GPL(blkcg_root);
45
46	struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css;
47
48	static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
49
50	static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */
51
52	static bool blkcg_policy_enabled(struct request_queue *q,
53	const struct blkcg_policy *pol)
54	{
55	return pol && test_bit(pol->plid, q->blkcg_pols);
56	}
57
58	/**
59	* blkg_free - free a blkg
60	* @blkg: blkg to free
61	*
62	* Free @blkg which may be partially allocated.
63	*/
64	static void blkg_free(struct blkcg_gq *blkg)
65	{
66	int i;
67
68	if (!blkg)
69	return;
70
71	for (i = 0; i < BLKCG_MAX_POLS; i++)
72	if (blkg->pd[i])
73	blkcg_policy[i]->pd_free_fn(blkg->pd[i]);
74
75	if (blkg->blkcg != &blkcg_root)
76	blk_exit_rl(&blkg->rl);
77
78	blkg_rwstat_exit(&blkg->stat_ios);
79	blkg_rwstat_exit(&blkg->stat_bytes);
80	kfree(blkg);
81	}
82
83	/**
84	* blkg_alloc - allocate a blkg
85	* @blkcg: block cgroup the new blkg is associated with
86	* @q: request_queue the new blkg is associated with
87	* @gfp_mask: allocation mask to use
88	*
89	* Allocate a new blkg assocating @blkcg and @q.
90	*/
91	static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
92	gfp_t gfp_mask)
93	{
94	struct blkcg_gq *blkg;
95	int i;
96
97	/* alloc and init base part */
98	blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node);
99	if (!blkg)
100	return NULL;
101
102	if (blkg_rwstat_init(&blkg->stat_bytes, gfp_mask) ||
103	blkg_rwstat_init(&blkg->stat_ios, gfp_mask))
104	goto err_free;
105
106	blkg->q = q;
107	INIT_LIST_HEAD(&blkg->q_node);
108	blkg->blkcg = blkcg;
109	atomic_set(&blkg->refcnt, 1);
110
111	/* root blkg uses @q->root_rl, init rl only for !root blkgs */
112	if (blkcg != &blkcg_root) {
113	if (blk_init_rl(&blkg->rl, q, gfp_mask))
114	goto err_free;
115	blkg->rl.blkg = blkg;
116	}
117
118	for (i = 0; i < BLKCG_MAX_POLS; i++) {
119	struct blkcg_policy *pol = blkcg_policy[i];
120	struct blkg_policy_data *pd;
121
122	if (!blkcg_policy_enabled(q, pol))
123	continue;
124
125	/* alloc per-policy data and attach it to blkg */
126	pd = pol->pd_alloc_fn(gfp_mask, q->node);
127	if (!pd)
128	goto err_free;
129
130	blkg->pd[i] = pd;
131	pd->blkg = blkg;
132	pd->plid = i;
133	}
134
135	return blkg;
136
137	err_free:
138	blkg_free(blkg);
139	return NULL;
140	}
141
142	struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
143	struct request_queue *q, bool update_hint)
144	{
145	struct blkcg_gq *blkg;
146
147	/*
148	* Hint didn't match. Look up from the radix tree. Note that the
149	* hint can only be updated under queue_lock as otherwise @blkg
150	* could have already been removed from blkg_tree. The caller is
151	* responsible for grabbing queue_lock if @update_hint.
152	*/
153	blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
154	if (blkg && blkg->q == q) {
155	if (update_hint) {
156	lockdep_assert_held(q->queue_lock);
157	rcu_assign_pointer(blkcg->blkg_hint, blkg);
158	}
159	return blkg;
160	}
161
162	return NULL;
163	}
164	EXPORT_SYMBOL_GPL(blkg_lookup_slowpath);
165
166	/*
167	* If @new_blkg is %NULL, this function tries to allocate a new one as
168	* necessary using %GFP_NOWAIT. @new_blkg is always consumed on return.
169	*/
170	static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
171	struct request_queue *q,
172	struct blkcg_gq *new_blkg)
173	{
174	struct blkcg_gq *blkg;
175	struct bdi_writeback_congested *wb_congested;
176	int i, ret;
177
178	WARN_ON_ONCE(!rcu_read_lock_held());
179	lockdep_assert_held(q->queue_lock);
180
181	/* blkg holds a reference to blkcg */
182	if (!css_tryget_online(&blkcg->css)) {
183	ret = -ENODEV;
184	goto err_free_blkg;
185	}
186
187	wb_congested = wb_congested_get_create(&q->backing_dev_info,
188	blkcg->css.id,
189	GFP_NOWAIT | __GFP_NOWARN);
190	if (!wb_congested) {
191	ret = -ENOMEM;
192	goto err_put_css;
193	}
194
195	/* allocate */
196	if (!new_blkg) {
197	new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT | __GFP_NOWARN);
198	if (unlikely(!new_blkg)) {
199	ret = -ENOMEM;
200	goto err_put_congested;
201	}
202	}
203	blkg = new_blkg;
204	blkg->wb_congested = wb_congested;
205
206	/* link parent */
207	if (blkcg_parent(blkcg)) {
208	blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false);
209	if (WARN_ON_ONCE(!blkg->parent)) {
210	ret = -ENODEV;
211	goto err_put_congested;
212	}
213	blkg_get(blkg->parent);
214	}
215
216	/* invoke per-policy init */
217	for (i = 0; i < BLKCG_MAX_POLS; i++) {
218	struct blkcg_policy *pol = blkcg_policy[i];
219
220	if (blkg->pd[i] && pol->pd_init_fn)
221	pol->pd_init_fn(blkg->pd[i]);
222	}
223
224	/* insert */
225	spin_lock(&blkcg->lock);
226	ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
227	if (likely(!ret)) {
228	hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
229	list_add(&blkg->q_node, &q->blkg_list);
230
231	for (i = 0; i < BLKCG_MAX_POLS; i++) {
232	struct blkcg_policy *pol = blkcg_policy[i];
233
234	if (blkg->pd[i] && pol->pd_online_fn)
235	pol->pd_online_fn(blkg->pd[i]);
236	}
237	}
238	blkg->online = true;
239	spin_unlock(&blkcg->lock);
240
241	if (!ret)
242	return blkg;
243
244	/* @blkg failed fully initialized, use the usual release path */
245	blkg_put(blkg);
246	return ERR_PTR(ret);
247
248	err_put_congested:
249	wb_congested_put(wb_congested);
250	err_put_css:
251	css_put(&blkcg->css);
252	err_free_blkg:
253	blkg_free(new_blkg);
254	return ERR_PTR(ret);
255	}
256
257	/**
258	* blkg_lookup_create - lookup blkg, try to create one if not there
259	* @blkcg: blkcg of interest
260	* @q: request_queue of interest
261	*
262	* Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
263	* create one. blkg creation is performed recursively from blkcg_root such
264	* that all non-root blkg's have access to the parent blkg. This function
265	* should be called under RCU read lock and @q->queue_lock.
266	*
267	* Returns pointer to the looked up or created blkg on success, ERR_PTR()
268	* value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
269	* dead and bypassing, returns ERR_PTR(-EBUSY).
270	*/
271	struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
272	struct request_queue *q)
273	{
274	struct blkcg_gq *blkg;
275
276	WARN_ON_ONCE(!rcu_read_lock_held());
277	lockdep_assert_held(q->queue_lock);
278
279	/*
280	* This could be the first entry point of blkcg implementation and
281	* we shouldn't allow anything to go through for a bypassing queue.
282	*/
283	if (unlikely(blk_queue_bypass(q)))
284	return ERR_PTR(blk_queue_dying(q) ? -ENODEV : -EBUSY);
285
286	blkg = __blkg_lookup(blkcg, q, true);
287	if (blkg)
288	return blkg;
289
290	/*
291	* Create blkgs walking down from blkcg_root to @blkcg, so that all
292	* non-root blkgs have access to their parents.
293	*/
294	while (true) {
295	struct blkcg *pos = blkcg;
296	struct blkcg *parent = blkcg_parent(blkcg);
297
298	while (parent && !__blkg_lookup(parent, q, false)) {
299	pos = parent;
300	parent = blkcg_parent(parent);
301	}
302
303	blkg = blkg_create(pos, q, NULL);
304	if (pos == blkcg || IS_ERR(blkg))
305	return blkg;
306	}
307	}
308
309	static void blkg_destroy(struct blkcg_gq *blkg)
310	{
311	struct blkcg *blkcg = blkg->blkcg;
312	struct blkcg_gq *parent = blkg->parent;
313	int i;
314
315	lockdep_assert_held(blkg->q->queue_lock);
316	lockdep_assert_held(&blkcg->lock);
317
318	/* Something wrong if we are trying to remove same group twice */
319	WARN_ON_ONCE(list_empty(&blkg->q_node));
320	WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
321
322	for (i = 0; i < BLKCG_MAX_POLS; i++) {
323	struct blkcg_policy *pol = blkcg_policy[i];
324
325	if (blkg->pd[i] && pol->pd_offline_fn)
326	pol->pd_offline_fn(blkg->pd[i]);
327	}
328
329	if (parent) {
330	blkg_rwstat_add_aux(&parent->stat_bytes, &blkg->stat_bytes);
331	blkg_rwstat_add_aux(&parent->stat_ios, &blkg->stat_ios);
332	}
333
334	blkg->online = false;
335
336	radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
337	list_del_init(&blkg->q_node);
338	hlist_del_init_rcu(&blkg->blkcg_node);
339
340	/*
341	* Both setting lookup hint to and clearing it from @blkg are done
342	* under queue_lock. If it's not pointing to @blkg now, it never
343	* will. Hint assignment itself can race safely.
344	*/
345	if (rcu_access_pointer(blkcg->blkg_hint) == blkg)
346	rcu_assign_pointer(blkcg->blkg_hint, NULL);
347
348	/*
349	* Put the reference taken at the time of creation so that when all
350	* queues are gone, group can be destroyed.
351	*/
352	blkg_put(blkg);
353	}
354
355	/**
356	* blkg_destroy_all - destroy all blkgs associated with a request_queue
357	* @q: request_queue of interest
358	*
359	* Destroy all blkgs associated with @q.
360	*/
361	static void blkg_destroy_all(struct request_queue *q)
362	{
363	struct blkcg_gq *blkg, *n;
364
365	lockdep_assert_held(q->queue_lock);
366
367	list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
368	struct blkcg *blkcg = blkg->blkcg;
369
370	spin_lock(&blkcg->lock);
371	blkg_destroy(blkg);
372	spin_unlock(&blkcg->lock);
373	}
374
375	q->root_blkg = NULL;
376	q->root_rl.blkg = NULL;
377	}
378
379	/*
380	* A group is RCU protected, but having an rcu lock does not mean that one
381	* can access all the fields of blkg and assume these are valid. For
382	* example, don't try to follow throtl_data and request queue links.
383	*
384	* Having a reference to blkg under an rcu allows accesses to only values
385	* local to groups like group stats and group rate limits.
386	*/
387	void __blkg_release_rcu(struct rcu_head *rcu_head)
388	{
389	struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head);
390
391	/* release the blkcg and parent blkg refs this blkg has been holding */
392	css_put(&blkg->blkcg->css);
393	if (blkg->parent)
394	blkg_put(blkg->parent);
395
396	wb_congested_put(blkg->wb_congested);
397
398	blkg_free(blkg);
399	}
400	EXPORT_SYMBOL_GPL(__blkg_release_rcu);
401
402	/*
403	* The next function used by blk_queue_for_each_rl(). It's a bit tricky
404	* because the root blkg uses @q->root_rl instead of its own rl.
405	*/
406	struct request_list *__blk_queue_next_rl(struct request_list *rl,
407	struct request_queue *q)
408	{
409	struct list_head *ent;
410	struct blkcg_gq *blkg;
411
412	/*
413	* Determine the current blkg list_head. The first entry is
414	* root_rl which is off @q->blkg_list and mapped to the head.
415	*/
416	if (rl == &q->root_rl) {
417	ent = &q->blkg_list;
418	/* There are no more block groups, hence no request lists */
419	if (list_empty(ent))
420	return NULL;
421	} else {
422	blkg = container_of(rl, struct blkcg_gq, rl);
423	ent = &blkg->q_node;
424	}
425
426	/* walk to the next list_head, skip root blkcg */
427	ent = ent->next;
428	if (ent == &q->root_blkg->q_node)
429	ent = ent->next;
430	if (ent == &q->blkg_list)
431	return NULL;
432
433	blkg = container_of(ent, struct blkcg_gq, q_node);
434	return &blkg->rl;
435	}
436
437	static int blkcg_reset_stats(struct cgroup_subsys_state *css,
438	struct cftype *cftype, u64 val)
439	{
440	struct blkcg *blkcg = css_to_blkcg(css);
441	struct blkcg_gq *blkg;
442	int i;
443
444	mutex_lock(&blkcg_pol_mutex);
445	spin_lock_irq(&blkcg->lock);
446
447	/*
448	* Note that stat reset is racy - it doesn't synchronize against
449	* stat updates. This is a debug feature which shouldn't exist
450	* anyway. If you get hit by a race, retry.
451	*/
452	hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
453	blkg_rwstat_reset(&blkg->stat_bytes);
454	blkg_rwstat_reset(&blkg->stat_ios);
455
456	for (i = 0; i < BLKCG_MAX_POLS; i++) {
457	struct blkcg_policy *pol = blkcg_policy[i];
458
459	if (blkg->pd[i] && pol->pd_reset_stats_fn)
460	pol->pd_reset_stats_fn(blkg->pd[i]);
461	}
462	}
463
464	spin_unlock_irq(&blkcg->lock);
465	mutex_unlock(&blkcg_pol_mutex);
466	return 0;
467	}
468
469	const char *blkg_dev_name(struct blkcg_gq *blkg)
470	{
471	/* some drivers (floppy) instantiate a queue w/o disk registered */
472	if (blkg->q->backing_dev_info.dev)
473	return dev_name(blkg->q->backing_dev_info.dev);
474	return NULL;
475	}
476	EXPORT_SYMBOL_GPL(blkg_dev_name);
477
478	/**
479	* blkcg_print_blkgs - helper for printing per-blkg data
480	* @sf: seq_file to print to
481	* @blkcg: blkcg of interest
482	* @prfill: fill function to print out a blkg
483	* @pol: policy in question
484	* @data: data to be passed to @prfill
485	* @show_total: to print out sum of prfill return values or not
486	*
487	* This function invokes @prfill on each blkg of @blkcg if pd for the
488	* policy specified by @pol exists. @prfill is invoked with @sf, the
489	* policy data and @data and the matching queue lock held. If @show_total
490	* is %true, the sum of the return values from @prfill is printed with
491	* "Total" label at the end.
492	*
493	* This is to be used to construct print functions for
494	* cftype->read_seq_string method.
495	*/
496	void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
497	u64 (*prfill)(struct seq_file *,
498	struct blkg_policy_data *, int),
499	const struct blkcg_policy *pol, int data,
500	bool show_total)
501	{
502	struct blkcg_gq *blkg;
503	u64 total = 0;
504
505	rcu_read_lock();
506	hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
507	spin_lock_irq(blkg->q->queue_lock);
508	if (blkcg_policy_enabled(blkg->q, pol))
509	total += prfill(sf, blkg->pd[pol->plid], data);
510	spin_unlock_irq(blkg->q->queue_lock);
511	}
512	rcu_read_unlock();
513
514	if (show_total)
515	seq_printf(sf, "Total %llu\n", (unsigned long long)total);
516	}
517	EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
518
519	/**
520	* __blkg_prfill_u64 - prfill helper for a single u64 value
521	* @sf: seq_file to print to
522	* @pd: policy private data of interest
523	* @v: value to print
524	*
525	* Print @v to @sf for the device assocaited with @pd.
526	*/
527	u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
528	{
529	const char *dname = blkg_dev_name(pd->blkg);
530
531	if (!dname)
532	return 0;
533
534	seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
535	return v;
536	}
537	EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
538
539	/**
540	* __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
541	* @sf: seq_file to print to
542	* @pd: policy private data of interest
543	* @rwstat: rwstat to print
544	*
545	* Print @rwstat to @sf for the device assocaited with @pd.
546	*/
547	u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
548	const struct blkg_rwstat *rwstat)
549	{
550	static const char *rwstr[] = {
551	[BLKG_RWSTAT_READ] = "Read",
552	[BLKG_RWSTAT_WRITE] = "Write",
553	[BLKG_RWSTAT_SYNC] = "Sync",
554	[BLKG_RWSTAT_ASYNC] = "Async",
555	};
556	const char *dname = blkg_dev_name(pd->blkg);
557	u64 v;
558	int i;
559
560	if (!dname)
561	return 0;
562
563	for (i = 0; i < BLKG_RWSTAT_NR; i++)
564	seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
565	(unsigned long long)atomic64_read(&rwstat->aux_cnt[i]));
566
567	v = atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_READ]) +
568	atomic64_read(&rwstat->aux_cnt[BLKG_RWSTAT_WRITE]);
569	seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
570	return v;
571	}
572	EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
573
574	/**
575	* blkg_prfill_stat - prfill callback for blkg_stat
576	* @sf: seq_file to print to
577	* @pd: policy private data of interest
578	* @off: offset to the blkg_stat in @pd
579	*
580	* prfill callback for printing a blkg_stat.
581	*/
582	u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
583	{
584	return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
585	}
586	EXPORT_SYMBOL_GPL(blkg_prfill_stat);
587
588	/**
589	* blkg_prfill_rwstat - prfill callback for blkg_rwstat
590	* @sf: seq_file to print to
591	* @pd: policy private data of interest
592	* @off: offset to the blkg_rwstat in @pd
593	*
594	* prfill callback for printing a blkg_rwstat.
595	*/
596	u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
597	int off)
598	{
599	struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
600
601	return __blkg_prfill_rwstat(sf, pd, &rwstat);
602	}
603	EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
604
605	static u64 blkg_prfill_rwstat_field(struct seq_file *sf,
606	struct blkg_policy_data *pd, int off)
607	{
608	struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd->blkg + off);
609
610	return __blkg_prfill_rwstat(sf, pd, &rwstat);
611	}
612
613	/**
614	* blkg_print_stat_bytes - seq_show callback for blkg->stat_bytes
615	* @sf: seq_file to print to
616	* @v: unused
617	*
618	* To be used as cftype->seq_show to print blkg->stat_bytes.
619	* cftype->private must be set to the blkcg_policy.
620	*/
621	int blkg_print_stat_bytes(struct seq_file *sf, void *v)
622	{
623	blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
624	blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private,
625	offsetof(struct blkcg_gq, stat_bytes), true);
626	return 0;
627	}
628	EXPORT_SYMBOL_GPL(blkg_print_stat_bytes);
629
630	/**
631	* blkg_print_stat_bytes - seq_show callback for blkg->stat_ios
632	* @sf: seq_file to print to
633	* @v: unused
634	*
635	* To be used as cftype->seq_show to print blkg->stat_ios. cftype->private
636	* must be set to the blkcg_policy.
637	*/
638	int blkg_print_stat_ios(struct seq_file *sf, void *v)
639	{
640	blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
641	blkg_prfill_rwstat_field, (void *)seq_cft(sf)->private,
642	offsetof(struct blkcg_gq, stat_ios), true);
643	return 0;
644	}
645	EXPORT_SYMBOL_GPL(blkg_print_stat_ios);
646
647	static u64 blkg_prfill_rwstat_field_recursive(struct seq_file *sf,
648	struct blkg_policy_data *pd,
649	int off)
650	{
651	struct blkg_rwstat rwstat = blkg_rwstat_recursive_sum(pd->blkg,
652	NULL, off);
653	return __blkg_prfill_rwstat(sf, pd, &rwstat);
654	}
655
656	/**
657	* blkg_print_stat_bytes_recursive - recursive version of blkg_print_stat_bytes
658	* @sf: seq_file to print to
659	* @v: unused
660	*/
661	int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v)
662	{
663	blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
664	blkg_prfill_rwstat_field_recursive,
665	(void *)seq_cft(sf)->private,
666	offsetof(struct blkcg_gq, stat_bytes), true);
667	return 0;
668	}
669	EXPORT_SYMBOL_GPL(blkg_print_stat_bytes_recursive);
670
671	/**
672	* blkg_print_stat_ios_recursive - recursive version of blkg_print_stat_ios
673	* @sf: seq_file to print to
674	* @v: unused
675	*/
676	int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v)
677	{
678	blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
679	blkg_prfill_rwstat_field_recursive,
680	(void *)seq_cft(sf)->private,
681	offsetof(struct blkcg_gq, stat_ios), true);
682	return 0;
683	}
684	EXPORT_SYMBOL_GPL(blkg_print_stat_ios_recursive);
685
686	/**
687	* blkg_stat_recursive_sum - collect hierarchical blkg_stat
688	* @blkg: blkg of interest
689	* @pol: blkcg_policy which contains the blkg_stat
690	* @off: offset to the blkg_stat in blkg_policy_data or @blkg
691	*
692	* Collect the blkg_stat specified by @blkg, @pol and @off and all its
693	* online descendants and their aux counts. The caller must be holding the
694	* queue lock for online tests.
695	*
696	* If @pol is NULL, blkg_stat is at @off bytes into @blkg; otherwise, it is
697	* at @off bytes into @blkg's blkg_policy_data of the policy.
698	*/
699	u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg,
700	struct blkcg_policy *pol, int off)
701	{
702	struct blkcg_gq *pos_blkg;
703	struct cgroup_subsys_state *pos_css;
704	u64 sum = 0;
705
706	lockdep_assert_held(blkg->q->queue_lock);
707
708	rcu_read_lock();
709	blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
710	struct blkg_stat *stat;
711
712	if (!pos_blkg->online)
713	continue;
714
715	if (pol)
716	stat = (void *)blkg_to_pd(pos_blkg, pol) + off;
717	else
718	stat = (void *)blkg + off;
719
720	sum += blkg_stat_read(stat) + atomic64_read(&stat->aux_cnt);
721	}
722	rcu_read_unlock();
723
724	return sum;
725	}
726	EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum);
727
728	/**
729	* blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
730	* @blkg: blkg of interest
731	* @pol: blkcg_policy which contains the blkg_rwstat
732	* @off: offset to the blkg_rwstat in blkg_policy_data or @blkg
733	*
734	* Collect the blkg_rwstat specified by @blkg, @pol and @off and all its
735	* online descendants and their aux counts. The caller must be holding the
736	* queue lock for online tests.
737	*
738	* If @pol is NULL, blkg_rwstat is at @off bytes into @blkg; otherwise, it
739	* is at @off bytes into @blkg's blkg_policy_data of the policy.
740	*/
741	struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg,
742	struct blkcg_policy *pol, int off)
743	{
744	struct blkcg_gq *pos_blkg;
745	struct cgroup_subsys_state *pos_css;
746	struct blkg_rwstat sum = { };
747	int i;
748
749	lockdep_assert_held(blkg->q->queue_lock);
750
751	rcu_read_lock();
752	blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
753	struct blkg_rwstat *rwstat;
754
755	if (!pos_blkg->online)
756	continue;
757
758	if (pol)
759	rwstat = (void *)blkg_to_pd(pos_blkg, pol) + off;
760	else
761	rwstat = (void *)pos_blkg + off;
762
763	for (i = 0; i < BLKG_RWSTAT_NR; i++)
764	atomic64_add(atomic64_read(&rwstat->aux_cnt[i]) +
765	percpu_counter_sum_positive(&rwstat->cpu_cnt[i]),
766	&sum.aux_cnt[i]);
767	}
768	rcu_read_unlock();
769
770	return sum;
771	}
772	EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
773
774	/**
775	* blkg_conf_prep - parse and prepare for per-blkg config update
776	* @blkcg: target block cgroup
777	* @pol: target policy
778	* @input: input string
779	* @ctx: blkg_conf_ctx to be filled
780	*
781	* Parse per-blkg config update from @input and initialize @ctx with the
782	* result. @ctx->blkg points to the blkg to be updated and @ctx->body the
783	* part of @input following MAJ:MIN. This function returns with RCU read
784	* lock and queue lock held and must be paired with blkg_conf_finish().
785	*/
786	int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
787	char *input, struct blkg_conf_ctx *ctx)
788	__acquires(rcu) __acquires(disk->queue->queue_lock)
789	{
790	struct gendisk *disk;
791	struct blkcg_gq *blkg;
792	struct module *owner;
793	unsigned int major, minor;
794	int key_len, part, ret;
795	char *body;
796
797	if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2)
798	return -EINVAL;
799
800	body = input + key_len;
801	if (!isspace(*body))
802	return -EINVAL;
803	body = skip_spaces(body);
804
805	disk = get_gendisk(MKDEV(major, minor), &part);
806	if (!disk)
807	return -ENODEV;
808	if (part) {
809	owner = disk->fops->owner;
810	put_disk(disk);
811	module_put(owner);
812	return -ENODEV;
813	}
814
815	rcu_read_lock();
816	spin_lock_irq(disk->queue->queue_lock);
817
818	if (blkcg_policy_enabled(disk->queue, pol))
819	blkg = blkg_lookup_create(blkcg, disk->queue);
820	else
821	blkg = ERR_PTR(-EOPNOTSUPP);
822
823	if (IS_ERR(blkg)) {
824	ret = PTR_ERR(blkg);
825	rcu_read_unlock();
826	spin_unlock_irq(disk->queue->queue_lock);
827	owner = disk->fops->owner;
828	put_disk(disk);
829	module_put(owner);
830	/*
831	* If queue was bypassing, we should retry. Do so after a
832	* short msleep(). It isn't strictly necessary but queue
833	* can be bypassing for some time and it's always nice to
834	* avoid busy looping.
835	*/
836	if (ret == -EBUSY) {
837	msleep(10);
838	ret = restart_syscall();
839	}
840	return ret;
841	}
842
843	ctx->disk = disk;
844	ctx->blkg = blkg;
845	ctx->body = body;
846	return 0;
847	}
848	EXPORT_SYMBOL_GPL(blkg_conf_prep);
849
850	/**
851	* blkg_conf_finish - finish up per-blkg config update
852	* @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
853	*
854	* Finish up after per-blkg config update. This function must be paired
855	* with blkg_conf_prep().
856	*/
857	void blkg_conf_finish(struct blkg_conf_ctx *ctx)
858	__releases(ctx->disk->queue->queue_lock) __releases(rcu)
859	{
860	struct module *owner;
861
862	spin_unlock_irq(ctx->disk->queue->queue_lock);
863	rcu_read_unlock();
864	owner = ctx->disk->fops->owner;
865	put_disk(ctx->disk);
866	module_put(owner);
867	}
868	EXPORT_SYMBOL_GPL(blkg_conf_finish);
869
870	static int blkcg_print_stat(struct seq_file *sf, void *v)
871	{
872	struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
873	struct blkcg_gq *blkg;
874
875	rcu_read_lock();
876
877	hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
878	const char *dname;
879	struct blkg_rwstat rwstat;
880	u64 rbytes, wbytes, rios, wios;
881
882	dname = blkg_dev_name(blkg);
883	if (!dname)
884	continue;
885
886	spin_lock_irq(blkg->q->queue_lock);
887
888	rwstat = blkg_rwstat_recursive_sum(blkg, NULL,
889	offsetof(struct blkcg_gq, stat_bytes));
890	rbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]);
891	wbytes = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]);
892
893	rwstat = blkg_rwstat_recursive_sum(blkg, NULL,
894	offsetof(struct blkcg_gq, stat_ios));
895	rios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_READ]);
896	wios = atomic64_read(&rwstat.aux_cnt[BLKG_RWSTAT_WRITE]);
897
898	spin_unlock_irq(blkg->q->queue_lock);
899
900	if (rbytes || wbytes || rios || wios)
901	seq_printf(sf, "%s rbytes=%llu wbytes=%llu rios=%llu wios=%llu\n",
902	dname, rbytes, wbytes, rios, wios);
903	}
904
905	rcu_read_unlock();
906	return 0;
907	}
908
909	static struct cftype blkcg_files[] = {
910	{
911	.name = "stat",
912	.flags = CFTYPE_NOT_ON_ROOT,
913	.seq_show = blkcg_print_stat,
914	},
915	{ } /* terminate */
916	};
917
918	static struct cftype blkcg_legacy_files[] = {
919	{
920	.name = "reset_stats",
921	.write_u64 = blkcg_reset_stats,
922	},
923	{ } /* terminate */
924	};
925
926	/**
927	* blkcg_css_offline - cgroup css_offline callback
928	* @css: css of interest
929	*
930	* This function is called when @css is about to go away and responsible
931	* for shooting down all blkgs associated with @css. blkgs should be
932	* removed while holding both q and blkcg locks. As blkcg lock is nested
933	* inside q lock, this function performs reverse double lock dancing.
934	*
935	* This is the blkcg counterpart of ioc_release_fn().
936	*/
937	static void blkcg_css_offline(struct cgroup_subsys_state *css)
938	{
939	struct blkcg *blkcg = css_to_blkcg(css);
940
941	spin_lock_irq(&blkcg->lock);
942
943	while (!hlist_empty(&blkcg->blkg_list)) {
944	struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
945	struct blkcg_gq, blkcg_node);
946	struct request_queue *q = blkg->q;
947
948	if (spin_trylock(q->queue_lock)) {
949	blkg_destroy(blkg);
950	spin_unlock(q->queue_lock);
951	} else {
952	spin_unlock_irq(&blkcg->lock);
953	cpu_relax();
954	spin_lock_irq(&blkcg->lock);
955	}
956	}
957
958	spin_unlock_irq(&blkcg->lock);
959
960	wb_blkcg_offline(blkcg);
961	}
962
963	static void blkcg_css_free(struct cgroup_subsys_state *css)
964	{
965	struct blkcg *blkcg = css_to_blkcg(css);
966	int i;
967
968	mutex_lock(&blkcg_pol_mutex);
969
970	list_del(&blkcg->all_blkcgs_node);
971
972	for (i = 0; i < BLKCG_MAX_POLS; i++)
973	if (blkcg->cpd[i])
974	blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
975
976	mutex_unlock(&blkcg_pol_mutex);
977
978	kfree(blkcg);
979	}
980
981	static struct cgroup_subsys_state *
982	blkcg_css_alloc(struct cgroup_subsys_state *parent_css)
983	{
984	struct blkcg *blkcg;
985	struct cgroup_subsys_state *ret;
986	int i;
987
988	mutex_lock(&blkcg_pol_mutex);
989
990	if (!parent_css) {
991	blkcg = &blkcg_root;
992	} else {
993	blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
994	if (!blkcg) {
995	ret = ERR_PTR(-ENOMEM);
996	goto free_blkcg;
997	}
998	}
999
1000	for (i = 0; i < BLKCG_MAX_POLS ; i++) {
1001	struct blkcg_policy *pol = blkcg_policy[i];
1002	struct blkcg_policy_data *cpd;
1003
1004	/*
1005	* If the policy hasn't been attached yet, wait for it
1006	* to be attached before doing anything else. Otherwise,
1007	* check if the policy requires any specific per-cgroup
1008	* data: if it does, allocate and initialize it.
1009	*/
1010	if (!pol || !pol->cpd_alloc_fn)
1011	continue;
1012
1013	cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1014	if (!cpd) {
1015	ret = ERR_PTR(-ENOMEM);
1016	goto free_pd_blkcg;
1017	}
1018	blkcg->cpd[i] = cpd;
1019	cpd->blkcg = blkcg;
1020	cpd->plid = i;
1021	if (pol->cpd_init_fn)
1022	pol->cpd_init_fn(cpd);
1023	}
1024
1025	spin_lock_init(&blkcg->lock);
1026	INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN);
1027	INIT_HLIST_HEAD(&blkcg->blkg_list);
1028	#ifdef CONFIG_CGROUP_WRITEBACK
1029	INIT_LIST_HEAD(&blkcg->cgwb_list);
1030	#endif
1031	list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs);
1032
1033	mutex_unlock(&blkcg_pol_mutex);
1034	return &blkcg->css;
1035
1036	free_pd_blkcg:
1037	for (i--; i >= 0; i--)
1038	if (blkcg->cpd[i])
1039	blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]);
1040	free_blkcg:
1041	kfree(blkcg);
1042	mutex_unlock(&blkcg_pol_mutex);
1043	return ret;
1044	}
1045
1046	/**
1047	* blkcg_init_queue - initialize blkcg part of request queue
1048	* @q: request_queue to initialize
1049	*
1050	* Called from blk_alloc_queue_node(). Responsible for initializing blkcg
1051	* part of new request_queue @q.
1052	*
1053	* RETURNS:
1054	* 0 on success, -errno on failure.
1055	*/
1056	int blkcg_init_queue(struct request_queue *q)
1057	{
1058	struct blkcg_gq *new_blkg, *blkg;
1059	bool preloaded;
1060	int ret;
1061
1062	new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
1063	if (!new_blkg)
1064	return -ENOMEM;
1065
1066	preloaded = !radix_tree_preload(GFP_KERNEL);
1067
1068	/*
1069	* Make sure the root blkg exists and count the existing blkgs. As
1070	* @q is bypassing at this point, blkg_lookup_create() can't be
1071	* used. Open code insertion.
1072	*/
1073	rcu_read_lock();
1074	spin_lock_irq(q->queue_lock);
1075	blkg = blkg_create(&blkcg_root, q, new_blkg);
1076	spin_unlock_irq(q->queue_lock);
1077	rcu_read_unlock();
1078
1079	if (preloaded)
1080	radix_tree_preload_end();
1081
1082	if (IS_ERR(blkg))
1083	return PTR_ERR(blkg);
1084
1085	q->root_blkg = blkg;
1086	q->root_rl.blkg = blkg;
1087
1088	ret = blk_throtl_init(q);
1089	if (ret) {
1090	spin_lock_irq(q->queue_lock);
1091	blkg_destroy_all(q);
1092	spin_unlock_irq(q->queue_lock);
1093	}
1094	return ret;
1095	}
1096
1097	/**
1098	* blkcg_drain_queue - drain blkcg part of request_queue
1099	* @q: request_queue to drain
1100	*
1101	* Called from blk_drain_queue(). Responsible for draining blkcg part.
1102	*/
1103	void blkcg_drain_queue(struct request_queue *q)
1104	{
1105	lockdep_assert_held(q->queue_lock);
1106
1107	/*
1108	* @q could be exiting and already have destroyed all blkgs as
1109	* indicated by NULL root_blkg. If so, don't confuse policies.
1110	*/
1111	if (!q->root_blkg)
1112	return;
1113
1114	blk_throtl_drain(q);
1115	}
1116
1117	/**
1118	* blkcg_exit_queue - exit and release blkcg part of request_queue
1119	* @q: request_queue being released
1120	*
1121	* Called from blk_release_queue(). Responsible for exiting blkcg part.
1122	*/
1123	void blkcg_exit_queue(struct request_queue *q)
1124	{
1125	spin_lock_irq(q->queue_lock);
1126	blkg_destroy_all(q);
1127	spin_unlock_irq(q->queue_lock);
1128
1129	blk_throtl_exit(q);
1130	}
1131
1132	/*
1133	* We cannot support shared io contexts, as we have no mean to support
1134	* two tasks with the same ioc in two different groups without major rework
1135	* of the main cic data structures. For now we allow a task to change
1136	* its cgroup only if it's the only owner of its ioc.
1137	*/
1138	static int blkcg_can_attach(struct cgroup_taskset *tset)
1139	{
1140	struct task_struct *task;
1141	struct cgroup_subsys_state *dst_css;
1142	struct io_context *ioc;
1143	int ret = 0;
1144
1145	/* task_lock() is needed to avoid races with exit_io_context() */
1146	cgroup_taskset_for_each(task, dst_css, tset) {
1147	task_lock(task);
1148	ioc = task->io_context;
1149	if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1150	ret = -EINVAL;
1151	task_unlock(task);
1152	if (ret)
1153	break;
1154	}
1155	return ret;
1156	}
1157
1158	static void blkcg_bind(struct cgroup_subsys_state *root_css)
1159	{
1160	int i;
1161
1162	mutex_lock(&blkcg_pol_mutex);
1163
1164	for (i = 0; i < BLKCG_MAX_POLS; i++) {
1165	struct blkcg_policy *pol = blkcg_policy[i];
1166	struct blkcg *blkcg;
1167
1168	if (!pol || !pol->cpd_bind_fn)
1169	continue;
1170
1171	list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node)
1172	if (blkcg->cpd[pol->plid])
1173	pol->cpd_bind_fn(blkcg->cpd[pol->plid]);
1174	}
1175	mutex_unlock(&blkcg_pol_mutex);
1176	}
1177
1178	struct cgroup_subsys io_cgrp_subsys = {
1179	.css_alloc = blkcg_css_alloc,
1180	.css_offline = blkcg_css_offline,
1181	.css_free = blkcg_css_free,
1182	.can_attach = blkcg_can_attach,
1183	.bind = blkcg_bind,
1184	.dfl_cftypes = blkcg_files,
1185	.legacy_cftypes = blkcg_legacy_files,
1186	.legacy_name = "blkio",
1187	#ifdef CONFIG_MEMCG
1188	/*
1189	* This ensures that, if available, memcg is automatically enabled
1190	* together on the default hierarchy so that the owner cgroup can
1191	* be retrieved from writeback pages.
1192	*/
1193	.depends_on = 1 << memory_cgrp_id,
1194	#endif
1195	};
1196	EXPORT_SYMBOL_GPL(io_cgrp_subsys);
1197
1198	/**
1199	* blkcg_activate_policy - activate a blkcg policy on a request_queue
1200	* @q: request_queue of interest
1201	* @pol: blkcg policy to activate
1202	*
1203	* Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
1204	* bypass mode to populate its blkgs with policy_data for @pol.
1205	*
1206	* Activation happens with @q bypassed, so nobody would be accessing blkgs
1207	* from IO path. Update of each blkg is protected by both queue and blkcg
1208	* locks so that holding either lock and testing blkcg_policy_enabled() is
1209	* always enough for dereferencing policy data.
1210	*
1211	* The caller is responsible for synchronizing [de]activations and policy
1212	* [un]registerations. Returns 0 on success, -errno on failure.
1213	*/
1214	int blkcg_activate_policy(struct request_queue *q,
1215	const struct blkcg_policy *pol)
1216	{
1217	struct blkg_policy_data *pd_prealloc = NULL;
1218	struct blkcg_gq *blkg;
1219	int ret;
1220
1221	if (blkcg_policy_enabled(q, pol))
1222	return 0;
1223
1224	blk_queue_bypass_start(q);
1225	pd_prealloc:
1226	if (!pd_prealloc) {
1227	pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q->node);
1228	if (!pd_prealloc) {
1229	ret = -ENOMEM;
1230	goto out_bypass_end;
1231	}
1232	}
1233
1234	spin_lock_irq(q->queue_lock);
1235
1236	list_for_each_entry(blkg, &q->blkg_list, q_node) {
1237	struct blkg_policy_data *pd;
1238
1239	if (blkg->pd[pol->plid])
1240	continue;
1241
1242	pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q->node);
1243	if (!pd)
1244	swap(pd, pd_prealloc);
1245	if (!pd) {
1246	spin_unlock_irq(q->queue_lock);
1247	goto pd_prealloc;
1248	}
1249
1250	blkg->pd[pol->plid] = pd;
1251	pd->blkg = blkg;
1252	pd->plid = pol->plid;
1253	if (pol->pd_init_fn)
1254	pol->pd_init_fn(pd);
1255	}
1256
1257	__set_bit(pol->plid, q->blkcg_pols);
1258	ret = 0;
1259
1260	spin_unlock_irq(q->queue_lock);
1261	out_bypass_end:
1262	blk_queue_bypass_end(q);
1263	if (pd_prealloc)
1264	pol->pd_free_fn(pd_prealloc);
1265	return ret;
1266	}
1267	EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1268
1269	/**
1270	* blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1271	* @q: request_queue of interest
1272	* @pol: blkcg policy to deactivate
1273	*
1274	* Deactivate @pol on @q. Follows the same synchronization rules as
1275	* blkcg_activate_policy().
1276	*/
1277	void blkcg_deactivate_policy(struct request_queue *q,
1278	const struct blkcg_policy *pol)
1279	{
1280	struct blkcg_gq *blkg;
1281
1282	if (!blkcg_policy_enabled(q, pol))
1283	return;
1284
1285	blk_queue_bypass_start(q);
1286	spin_lock_irq(q->queue_lock);
1287
1288	__clear_bit(pol->plid, q->blkcg_pols);
1289
1290	list_for_each_entry(blkg, &q->blkg_list, q_node) {
1291	/* grab blkcg lock too while removing @pd from @blkg */
1292	spin_lock(&blkg->blkcg->lock);
1293
1294	if (blkg->pd[pol->plid]) {
1295	if (pol->pd_offline_fn)
1296	pol->pd_offline_fn(blkg->pd[pol->plid]);
1297	pol->pd_free_fn(blkg->pd[pol->plid]);
1298	blkg->pd[pol->plid] = NULL;
1299	}
1300
1301	spin_unlock(&blkg->blkcg->lock);
1302	}
1303
1304	spin_unlock_irq(q->queue_lock);
1305	blk_queue_bypass_end(q);
1306	}
1307	EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1308
1309	/**
1310	* blkcg_policy_register - register a blkcg policy
1311	* @pol: blkcg policy to register
1312	*
1313	* Register @pol with blkcg core. Might sleep and @pol may be modified on
1314	* successful registration. Returns 0 on success and -errno on failure.
1315	*/
1316	int blkcg_policy_register(struct blkcg_policy *pol)
1317	{
1318	struct blkcg *blkcg;
1319	int i, ret;
1320
1321	mutex_lock(&blkcg_pol_register_mutex);
1322	mutex_lock(&blkcg_pol_mutex);
1323
1324	/* find an empty slot */
1325	ret = -ENOSPC;
1326	for (i = 0; i < BLKCG_MAX_POLS; i++)
1327	if (!blkcg_policy[i])
1328	break;
1329	if (i >= BLKCG_MAX_POLS)
1330	goto err_unlock;
1331
1332	/* register @pol */
1333	pol->plid = i;
1334	blkcg_policy[pol->plid] = pol;
1335
1336	/* allocate and install cpd's */
1337	if (pol->cpd_alloc_fn) {
1338	list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1339	struct blkcg_policy_data *cpd;
1340
1341	cpd = pol->cpd_alloc_fn(GFP_KERNEL);
1342	if (!cpd)
1343	goto err_free_cpds;
1344
1345	blkcg->cpd[pol->plid] = cpd;
1346	cpd->blkcg = blkcg;
1347	cpd->plid = pol->plid;
1348	pol->cpd_init_fn(cpd);
1349	}
1350	}
1351
1352	mutex_unlock(&blkcg_pol_mutex);
1353
1354	/* everything is in place, add intf files for the new policy */
1355	if (pol->dfl_cftypes)
1356	WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys,
1357	pol->dfl_cftypes));
1358	if (pol->legacy_cftypes)
1359	WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys,
1360	pol->legacy_cftypes));
1361	mutex_unlock(&blkcg_pol_register_mutex);
1362	return 0;
1363
1364	err_free_cpds:
1365	if (pol->cpd_alloc_fn) {
1366	list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1367	if (blkcg->cpd[pol->plid]) {
1368	pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1369	blkcg->cpd[pol->plid] = NULL;
1370	}
1371	}
1372	}
1373	blkcg_policy[pol->plid] = NULL;
1374	err_unlock:
1375	mutex_unlock(&blkcg_pol_mutex);
1376	mutex_unlock(&blkcg_pol_register_mutex);
1377	return ret;
1378	}
1379	EXPORT_SYMBOL_GPL(blkcg_policy_register);
1380
1381	/**
1382	* blkcg_policy_unregister - unregister a blkcg policy
1383	* @pol: blkcg policy to unregister
1384	*
1385	* Undo blkcg_policy_register(@pol). Might sleep.
1386	*/
1387	void blkcg_policy_unregister(struct blkcg_policy *pol)
1388	{
1389	struct blkcg *blkcg;
1390
1391	mutex_lock(&blkcg_pol_register_mutex);
1392
1393	if (WARN_ON(blkcg_policy[pol->plid] != pol))
1394	goto out_unlock;
1395
1396	/* kill the intf files first */
1397	if (pol->dfl_cftypes)
1398	cgroup_rm_cftypes(pol->dfl_cftypes);
1399	if (pol->legacy_cftypes)
1400	cgroup_rm_cftypes(pol->legacy_cftypes);
1401
1402	/* remove cpds and unregister */
1403	mutex_lock(&blkcg_pol_mutex);
1404
1405	if (pol->cpd_alloc_fn) {
1406	list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) {
1407	if (blkcg->cpd[pol->plid]) {
1408	pol->cpd_free_fn(blkcg->cpd[pol->plid]);
1409	blkcg->cpd[pol->plid] = NULL;
1410	}
1411	}
1412	}
1413	blkcg_policy[pol->plid] = NULL;
1414
1415	mutex_unlock(&blkcg_pol_mutex);
1416	out_unlock:
1417	mutex_unlock(&blkcg_pol_register_mutex);
1418	}
1419	EXPORT_SYMBOL_GPL(blkcg_policy_unregister);
1420