2 * Common Block IO controller cgroup interface
4 * Based on ideas and code from CFQ, CFS and BFQ:
5 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
7 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
8 * Paolo Valente <paolo.valente@unimore.it>
10 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
11 * Nauman Rafique <nauman@google.com>
13 #include <linux/ioprio.h>
14 #include <linux/kdev_t.h>
15 #include <linux/module.h>
16 #include <linux/err.h>
17 #include <linux/blkdev.h>
18 #include <linux/slab.h>
19 #include <linux/genhd.h>
20 #include <linux/delay.h>
21 #include <linux/atomic.h>
22 #include "blk-cgroup.h"
25 #define MAX_KEY_LEN 100
27 static DEFINE_SPINLOCK(blkio_list_lock);
28 static LIST_HEAD(blkio_list);
30 static DEFINE_MUTEX(all_q_mutex);
31 static LIST_HEAD(all_q_list);
33 struct blkio_cgroup blkio_root_cgroup = { .cfq_weight = 2 * CFQ_WEIGHT_DEFAULT };
34 EXPORT_SYMBOL_GPL(blkio_root_cgroup);
36 static struct blkio_policy_type *blkio_policy[BLKIO_NR_POLICIES];
38 struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup)
40 return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id),
41 struct blkio_cgroup, css);
43 EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup);
45 static struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk)
47 return container_of(task_subsys_state(tsk, blkio_subsys_id),
48 struct blkio_cgroup, css);
51 struct blkio_cgroup *bio_blkio_cgroup(struct bio *bio)
53 if (bio && bio->bi_css)
54 return container_of(bio->bi_css, struct blkio_cgroup, css);
55 return task_blkio_cgroup(current);
57 EXPORT_SYMBOL_GPL(bio_blkio_cgroup);
60 * blkg_free - free a blkg
63 * Free @blkg which may be partially allocated.
65 static void blkg_free(struct blkio_group *blkg)
72 for (i = 0; i < BLKIO_NR_POLICIES; i++) {
73 struct blkio_policy_type *pol = blkio_policy[i];
74 struct blkg_policy_data *pd = blkg->pd[i];
79 if (pol && pol->ops.blkio_exit_group_fn)
80 pol->ops.blkio_exit_group_fn(blkg);
89 * blkg_alloc - allocate a blkg
90 * @blkcg: block cgroup the new blkg is associated with
91 * @q: request_queue the new blkg is associated with
93 * Allocate a new blkg assocating @blkcg and @q.
95 static struct blkio_group *blkg_alloc(struct blkio_cgroup *blkcg,
96 struct request_queue *q)
98 struct blkio_group *blkg;
101 /* alloc and init base part */
102 blkg = kzalloc_node(sizeof(*blkg), GFP_ATOMIC, q->node);
107 INIT_LIST_HEAD(&blkg->q_node);
110 cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path));
112 for (i = 0; i < BLKIO_NR_POLICIES; i++) {
113 struct blkio_policy_type *pol = blkio_policy[i];
114 struct blkg_policy_data *pd;
119 /* alloc per-policy data and attach it to blkg */
120 pd = kzalloc_node(sizeof(*pd) + pol->pdata_size, GFP_ATOMIC,
131 /* invoke per-policy init */
132 for (i = 0; i < BLKIO_NR_POLICIES; i++) {
133 struct blkio_policy_type *pol = blkio_policy[i];
136 pol->ops.blkio_init_group_fn(blkg);
142 struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg,
143 struct request_queue *q,
145 __releases(q->queue_lock) __acquires(q->queue_lock)
147 struct blkio_group *blkg;
149 WARN_ON_ONCE(!rcu_read_lock_held());
150 lockdep_assert_held(q->queue_lock);
153 * This could be the first entry point of blkcg implementation and
154 * we shouldn't allow anything to go through for a bypassing queue.
155 * The following can be removed if blkg lookup is guaranteed to
156 * fail on a bypassing queue.
158 if (unlikely(blk_queue_bypass(q)) && !for_root)
159 return ERR_PTR(blk_queue_dead(q) ? -EINVAL : -EBUSY);
161 blkg = blkg_lookup(blkcg, q);
165 /* blkg holds a reference to blkcg */
166 if (!css_tryget(&blkcg->css))
167 return ERR_PTR(-EINVAL);
170 * Allocate and initialize.
172 blkg = blkg_alloc(blkcg, q);
174 /* did alloc fail? */
175 if (unlikely(!blkg)) {
176 blkg = ERR_PTR(-ENOMEM);
181 spin_lock(&blkcg->lock);
182 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
183 list_add(&blkg->q_node, &q->blkg_list);
184 spin_unlock(&blkcg->lock);
188 EXPORT_SYMBOL_GPL(blkg_lookup_create);
190 /* called under rcu_read_lock(). */
191 struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg,
192 struct request_queue *q)
194 struct blkio_group *blkg;
195 struct hlist_node *n;
197 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node)
202 EXPORT_SYMBOL_GPL(blkg_lookup);
204 static void blkg_destroy(struct blkio_group *blkg)
206 struct request_queue *q = blkg->q;
207 struct blkio_cgroup *blkcg = blkg->blkcg;
209 lockdep_assert_held(q->queue_lock);
210 lockdep_assert_held(&blkcg->lock);
212 /* Something wrong if we are trying to remove same group twice */
213 WARN_ON_ONCE(list_empty(&blkg->q_node));
214 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
215 list_del_init(&blkg->q_node);
216 hlist_del_init_rcu(&blkg->blkcg_node);
219 * Put the reference taken at the time of creation so that when all
220 * queues are gone, group can be destroyed.
226 * XXX: This updates blkg policy data in-place for root blkg, which is
227 * necessary across elevator switch and policy registration as root blkgs
228 * aren't shot down. This broken and racy implementation is temporary.
229 * Eventually, blkg shoot down will be replaced by proper in-place update.
231 void update_root_blkg_pd(struct request_queue *q, enum blkio_policy_id plid)
233 struct blkio_policy_type *pol = blkio_policy[plid];
234 struct blkio_group *blkg = blkg_lookup(&blkio_root_cgroup, q);
235 struct blkg_policy_data *pd;
240 kfree(blkg->pd[plid]);
241 blkg->pd[plid] = NULL;
246 pd = kzalloc(sizeof(*pd) + pol->pdata_size, GFP_KERNEL);
251 pol->ops.blkio_init_group_fn(blkg);
253 EXPORT_SYMBOL_GPL(update_root_blkg_pd);
256 * blkg_destroy_all - destroy all blkgs associated with a request_queue
257 * @q: request_queue of interest
258 * @destroy_root: whether to destroy root blkg or not
260 * Destroy blkgs associated with @q. If @destroy_root is %true, all are
261 * destroyed; otherwise, root blkg is left alone.
263 void blkg_destroy_all(struct request_queue *q, bool destroy_root)
265 struct blkio_group *blkg, *n;
267 spin_lock_irq(q->queue_lock);
269 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
270 struct blkio_cgroup *blkcg = blkg->blkcg;
273 if (!destroy_root && blkg->blkcg == &blkio_root_cgroup)
276 spin_lock(&blkcg->lock);
278 spin_unlock(&blkcg->lock);
281 spin_unlock_irq(q->queue_lock);
283 EXPORT_SYMBOL_GPL(blkg_destroy_all);
285 static void blkg_rcu_free(struct rcu_head *rcu_head)
287 blkg_free(container_of(rcu_head, struct blkio_group, rcu_head));
290 void __blkg_release(struct blkio_group *blkg)
292 /* release the extra blkcg reference this blkg has been holding */
293 css_put(&blkg->blkcg->css);
296 * A group is freed in rcu manner. But having an rcu lock does not
297 * mean that one can access all the fields of blkg and assume these
298 * are valid. For example, don't try to follow throtl_data and
299 * request queue links.
301 * Having a reference to blkg under an rcu allows acess to only
302 * values local to groups like group stats and group rate limits
304 call_rcu(&blkg->rcu_head, blkg_rcu_free);
306 EXPORT_SYMBOL_GPL(__blkg_release);
309 blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val)
311 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
312 struct blkio_group *blkg;
313 struct hlist_node *n;
315 spin_lock(&blkio_list_lock);
316 spin_lock_irq(&blkcg->lock);
319 * Note that stat reset is racy - it doesn't synchronize against
320 * stat updates. This is a debug feature which shouldn't exist
321 * anyway. If you get hit by a race, retry.
323 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
324 struct blkio_policy_type *pol;
326 list_for_each_entry(pol, &blkio_list, list)
327 if (pol->ops.blkio_reset_group_stats_fn)
328 pol->ops.blkio_reset_group_stats_fn(blkg);
331 spin_unlock_irq(&blkcg->lock);
332 spin_unlock(&blkio_list_lock);
336 static const char *blkg_dev_name(struct blkio_group *blkg)
338 /* some drivers (floppy) instantiate a queue w/o disk registered */
339 if (blkg->q->backing_dev_info.dev)
340 return dev_name(blkg->q->backing_dev_info.dev);
345 * blkcg_print_blkgs - helper for printing per-blkg data
346 * @sf: seq_file to print to
347 * @blkcg: blkcg of interest
348 * @prfill: fill function to print out a blkg
349 * @pol: policy in question
350 * @data: data to be passed to @prfill
351 * @show_total: to print out sum of prfill return values or not
353 * This function invokes @prfill on each blkg of @blkcg if pd for the
354 * policy specified by @pol exists. @prfill is invoked with @sf, the
355 * policy data and @data. If @show_total is %true, the sum of the return
356 * values from @prfill is printed with "Total" label at the end.
358 * This is to be used to construct print functions for
359 * cftype->read_seq_string method.
361 void blkcg_print_blkgs(struct seq_file *sf, struct blkio_cgroup *blkcg,
362 u64 (*prfill)(struct seq_file *, void *, int),
363 int pol, int data, bool show_total)
365 struct blkio_group *blkg;
366 struct hlist_node *n;
369 spin_lock_irq(&blkcg->lock);
370 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node)
372 total += prfill(sf, blkg->pd[pol]->pdata, data);
373 spin_unlock_irq(&blkcg->lock);
376 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
378 EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
381 * __blkg_prfill_u64 - prfill helper for a single u64 value
382 * @sf: seq_file to print to
383 * @pdata: policy private data of interest
386 * Print @v to @sf for the device assocaited with @pdata.
388 u64 __blkg_prfill_u64(struct seq_file *sf, void *pdata, u64 v)
390 const char *dname = blkg_dev_name(pdata_to_blkg(pdata));
395 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
398 EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
401 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
402 * @sf: seq_file to print to
403 * @pdata: policy private data of interest
404 * @rwstat: rwstat to print
406 * Print @rwstat to @sf for the device assocaited with @pdata.
408 u64 __blkg_prfill_rwstat(struct seq_file *sf, void *pdata,
409 const struct blkg_rwstat *rwstat)
411 static const char *rwstr[] = {
412 [BLKG_RWSTAT_READ] = "Read",
413 [BLKG_RWSTAT_WRITE] = "Write",
414 [BLKG_RWSTAT_SYNC] = "Sync",
415 [BLKG_RWSTAT_ASYNC] = "Async",
417 const char *dname = blkg_dev_name(pdata_to_blkg(pdata));
424 for (i = 0; i < BLKG_RWSTAT_NR; i++)
425 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
426 (unsigned long long)rwstat->cnt[i]);
428 v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE];
429 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
434 * blkg_prfill_stat - prfill callback for blkg_stat
435 * @sf: seq_file to print to
436 * @pdata: policy private data of interest
437 * @off: offset to the blkg_stat in @pdata
439 * prfill callback for printing a blkg_stat.
441 u64 blkg_prfill_stat(struct seq_file *sf, void *pdata, int off)
443 return __blkg_prfill_u64(sf, pdata, blkg_stat_read(pdata + off));
445 EXPORT_SYMBOL_GPL(blkg_prfill_stat);
448 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
449 * @sf: seq_file to print to
450 * @pdata: policy private data of interest
451 * @off: offset to the blkg_rwstat in @pdata
453 * prfill callback for printing a blkg_rwstat.
455 u64 blkg_prfill_rwstat(struct seq_file *sf, void *pdata, int off)
457 struct blkg_rwstat rwstat = blkg_rwstat_read(pdata + off);
459 return __blkg_prfill_rwstat(sf, pdata, &rwstat);
461 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
464 * blkg_conf_prep - parse and prepare for per-blkg config update
465 * @blkcg: target block cgroup
466 * @input: input string
467 * @ctx: blkg_conf_ctx to be filled
469 * Parse per-blkg config update from @input and initialize @ctx with the
470 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
471 * value. This function returns with RCU read locked and must be paired
472 * with blkg_conf_finish().
474 int blkg_conf_prep(struct blkio_cgroup *blkcg, const char *input,
475 struct blkg_conf_ctx *ctx)
478 struct gendisk *disk;
479 struct blkio_group *blkg;
480 unsigned int major, minor;
481 unsigned long long v;
484 if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3)
487 disk = get_gendisk(MKDEV(major, minor), &part);
493 spin_lock_irq(disk->queue->queue_lock);
494 blkg = blkg_lookup_create(blkcg, disk->queue, false);
495 spin_unlock_irq(disk->queue->queue_lock);
502 * If queue was bypassing, we should retry. Do so after a
503 * short msleep(). It isn't strictly necessary but queue
504 * can be bypassing for some time and it's always nice to
505 * avoid busy looping.
509 ret = restart_syscall();
519 EXPORT_SYMBOL_GPL(blkg_conf_prep);
522 * blkg_conf_finish - finish up per-blkg config update
523 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
525 * Finish up after per-blkg config update. This function must be paired
526 * with blkg_conf_prep().
528 void blkg_conf_finish(struct blkg_conf_ctx *ctx)
534 EXPORT_SYMBOL_GPL(blkg_conf_finish);
536 struct cftype blkio_files[] = {
538 .name = "reset_stats",
539 .write_u64 = blkiocg_reset_stats,
545 * blkiocg_pre_destroy - cgroup pre_destroy callback
546 * @cgroup: cgroup of interest
548 * This function is called when @cgroup is about to go away and responsible
549 * for shooting down all blkgs associated with @cgroup. blkgs should be
550 * removed while holding both q and blkcg locks. As blkcg lock is nested
551 * inside q lock, this function performs reverse double lock dancing.
553 * This is the blkcg counterpart of ioc_release_fn().
555 static int blkiocg_pre_destroy(struct cgroup *cgroup)
557 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
559 spin_lock_irq(&blkcg->lock);
561 while (!hlist_empty(&blkcg->blkg_list)) {
562 struct blkio_group *blkg = hlist_entry(blkcg->blkg_list.first,
563 struct blkio_group, blkcg_node);
564 struct request_queue *q = blkg->q;
566 if (spin_trylock(q->queue_lock)) {
568 spin_unlock(q->queue_lock);
570 spin_unlock_irq(&blkcg->lock);
572 spin_lock_irq(&blkcg->lock);
576 spin_unlock_irq(&blkcg->lock);
580 static void blkiocg_destroy(struct cgroup *cgroup)
582 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
584 if (blkcg != &blkio_root_cgroup)
588 static struct cgroup_subsys_state *blkiocg_create(struct cgroup *cgroup)
590 static atomic64_t id_seq = ATOMIC64_INIT(0);
591 struct blkio_cgroup *blkcg;
592 struct cgroup *parent = cgroup->parent;
595 blkcg = &blkio_root_cgroup;
599 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
601 return ERR_PTR(-ENOMEM);
603 blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT;
604 blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */
606 spin_lock_init(&blkcg->lock);
607 INIT_HLIST_HEAD(&blkcg->blkg_list);
613 * blkcg_init_queue - initialize blkcg part of request queue
614 * @q: request_queue to initialize
616 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
617 * part of new request_queue @q.
620 * 0 on success, -errno on failure.
622 int blkcg_init_queue(struct request_queue *q)
628 ret = blk_throtl_init(q);
632 mutex_lock(&all_q_mutex);
633 INIT_LIST_HEAD(&q->all_q_node);
634 list_add_tail(&q->all_q_node, &all_q_list);
635 mutex_unlock(&all_q_mutex);
641 * blkcg_drain_queue - drain blkcg part of request_queue
642 * @q: request_queue to drain
644 * Called from blk_drain_queue(). Responsible for draining blkcg part.
646 void blkcg_drain_queue(struct request_queue *q)
648 lockdep_assert_held(q->queue_lock);
654 * blkcg_exit_queue - exit and release blkcg part of request_queue
655 * @q: request_queue being released
657 * Called from blk_release_queue(). Responsible for exiting blkcg part.
659 void blkcg_exit_queue(struct request_queue *q)
661 mutex_lock(&all_q_mutex);
662 list_del_init(&q->all_q_node);
663 mutex_unlock(&all_q_mutex);
665 blkg_destroy_all(q, true);
671 * We cannot support shared io contexts, as we have no mean to support
672 * two tasks with the same ioc in two different groups without major rework
673 * of the main cic data structures. For now we allow a task to change
674 * its cgroup only if it's the only owner of its ioc.
676 static int blkiocg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
678 struct task_struct *task;
679 struct io_context *ioc;
682 /* task_lock() is needed to avoid races with exit_io_context() */
683 cgroup_taskset_for_each(task, cgrp, tset) {
685 ioc = task->io_context;
686 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
695 static void blkcg_bypass_start(void)
696 __acquires(&all_q_mutex)
698 struct request_queue *q;
700 mutex_lock(&all_q_mutex);
702 list_for_each_entry(q, &all_q_list, all_q_node) {
703 blk_queue_bypass_start(q);
704 blkg_destroy_all(q, false);
708 static void blkcg_bypass_end(void)
709 __releases(&all_q_mutex)
711 struct request_queue *q;
713 list_for_each_entry(q, &all_q_list, all_q_node)
714 blk_queue_bypass_end(q);
716 mutex_unlock(&all_q_mutex);
719 struct cgroup_subsys blkio_subsys = {
721 .create = blkiocg_create,
722 .can_attach = blkiocg_can_attach,
723 .pre_destroy = blkiocg_pre_destroy,
724 .destroy = blkiocg_destroy,
725 .subsys_id = blkio_subsys_id,
726 .base_cftypes = blkio_files,
727 .module = THIS_MODULE,
729 EXPORT_SYMBOL_GPL(blkio_subsys);
731 void blkio_policy_register(struct blkio_policy_type *blkiop)
733 struct request_queue *q;
735 blkcg_bypass_start();
736 spin_lock(&blkio_list_lock);
738 BUG_ON(blkio_policy[blkiop->plid]);
739 blkio_policy[blkiop->plid] = blkiop;
740 list_add_tail(&blkiop->list, &blkio_list);
742 spin_unlock(&blkio_list_lock);
743 list_for_each_entry(q, &all_q_list, all_q_node)
744 update_root_blkg_pd(q, blkiop->plid);
748 WARN_ON(cgroup_add_cftypes(&blkio_subsys, blkiop->cftypes));
750 EXPORT_SYMBOL_GPL(blkio_policy_register);
752 void blkio_policy_unregister(struct blkio_policy_type *blkiop)
754 struct request_queue *q;
757 cgroup_rm_cftypes(&blkio_subsys, blkiop->cftypes);
759 blkcg_bypass_start();
760 spin_lock(&blkio_list_lock);
762 BUG_ON(blkio_policy[blkiop->plid] != blkiop);
763 blkio_policy[blkiop->plid] = NULL;
764 list_del_init(&blkiop->list);
766 spin_unlock(&blkio_list_lock);
767 list_for_each_entry(q, &all_q_list, all_q_node)
768 update_root_blkg_pd(q, blkiop->plid);
771 EXPORT_SYMBOL_GPL(blkio_policy_unregister);