block, cgroup: implement policy-specific per-blkcg data

The block IO (blkio) controller enables the block layer to provide service
guarantees in a hierarchical fashion. Specifically, service guarantees
are provided by registered request-accounting policies. As of now, a
proportional-share and a throttling policy are available. They are
implemented, respectively, by the CFQ I/O scheduler and the blk-throttle
subsystem. Unfortunately, as for adding new policies, the current
implementation of the block IO controller is only halfway ready to allow
new policies to be plugged in. This commit provides a solution to make
the block IO controller fully ready to handle new policies.
In what follows, we first describe briefly the current state, and then
list the changes made by this commit.

The throttling policy does not need any per-cgroup information to perform
its task. In contrast, the proportional share policy uses, for each cgroup,
both the weight assigned by the user to the cgroup, and a set of dynamically-
computed weights, one for each device.

The first, user-defined weight is stored in the blkcg data structure: the
block IO controller allocates a private blkcg data structure for each
cgroup in the blkio cgroups hierarchy (regardless of which policy is active).
In other words, the block IO controller internally mirrors the blkio cgroups
with private blkcg data structures.

On the other hand, for each cgroup and device, the corresponding dynamically-
computed weight is maintained in the following, different way. For each device,
the block IO controller keeps a private blkcg_gq structure for each cgroup in
blkio. In other words, block IO also keeps one private mirror copy of the blkio
cgroups hierarchy for each device, made of blkcg_gq structures.
Each blkcg_gq structure keeps per-policy information in a generic array of
dynamically-allocated 'dedicated' data structures, one for each registered
policy (so currently the array contains two elements). To be inserted into the
generic array, each dedicated data structure embeds a generic blkg_policy_data
structure. Consider now the array contained in the blkcg_gq structure
corresponding to a given pair of cgroup and device: one of the elements
of the array contains the dedicated data structure for the proportional-share
policy, and this dedicated data structure contains the dynamically-computed
weight for that pair of cgroup and device.

The generic strategy adopted for storing per-policy data in blkcg_gq structures
is already capable of handling new policies, whereas the one adopted with blkcg
structures is not, because per-policy data are hard-coded in the blkcg
structures themselves (currently only data related to the proportional-
share policy).

This commit addresses the above issues through the following changes:
. It generalizes blkcg structures so that per-policy data are stored in the same
  way as in blkcg_gq structures.
  Specifically, it lets also the blkcg structure store per-policy data in a
  generic array of dynamically-allocated dedicated data structures. We will
  refer to these data structures as blkcg dedicated data structures, to
  distinguish them from the dedicated data structures inserted in the generic
  arrays kept by blkcg_gq structures.
  To allow blkcg dedicated data structures to be inserted in the generic array
  inside a blkcg structure, this commit also introduces a new blkcg_policy_data
  structure, which is the equivalent of blkg_policy_data for blkcg dedicated
  data structures.
. It adds to the blkcg_policy structure, i.e., to the descriptor of a policy, a
  cpd_size field and a cpd_init field, to be initialized by the policy with,
  respectively, the size of the blkcg dedicated data structures, and the
  address of a constructor function for blkcg dedicated data structures.
. It moves the CFQ-specific fields embedded in the blkcg data structure (i.e.,
  the fields related to the proportional-share policy), into a new blkcg
  dedicated data structure called cfq_group_data.

Signed-off-by: Paolo Valente <paolo.valente@unimore.it>
Signed-off-by: Arianna Avanzini <avanzini.arianna@gmail.com>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: Jens Axboe <axboe@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

3 files changed, 173 insertions, 29 deletions

diff --git a/block/blk-cgroup.c b/block/blk-cgroup.c
index 0ac817b..6e43fa3 100644
--- a/block/blk-cgroup.c
+++ b/block/blk-cgroup.c
@@ -9,6 +9,10 @@
  *
  * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
  * 	              Nauman Rafique <nauman@google.com>
+ *
+ * For policy-specific per-blkcg data:
+ * Copyright (C) 2015 Paolo Valente <paolo.valente@unimore.it>
+ *                    Arianna Avanzini <avanzini.arianna@gmail.com>
  */
 #include <linux/ioprio.h>
 #include <linux/kdev_t.h>
@@ -26,8 +30,7 @@
 
 static DEFINE_MUTEX(blkcg_pol_mutex);
 
-struct blkcg blkcg_root = { .cfq_weight = 2 * CFQ_WEIGHT_DEFAULT,
-			    .cfq_leaf_weight = 2 * CFQ_WEIGHT_DEFAULT, };
+struct blkcg blkcg_root;
 EXPORT_SYMBOL_GPL(blkcg_root);
 
 static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
@@ -823,6 +826,8 @@ static struct cgroup_subsys_state *
 blkcg_css_alloc(struct cgroup_subsys_state *parent_css)
 {
 	struct blkcg *blkcg;
+	struct cgroup_subsys_state *ret;
+	int i;
 
 	if (!parent_css) {
 		blkcg = &blkcg_root;
@@ -830,17 +835,49 @@ blkcg_css_alloc(struct cgroup_subsys_state *parent_css)
 	}
 
 	blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
-	if (!blkcg)
-		return ERR_PTR(-ENOMEM);
+	if (!blkcg) {
+		ret = ERR_PTR(-ENOMEM);
+		goto free_blkcg;
+	}
+
+	for (i = 0; i < BLKCG_MAX_POLS ; i++) {
+		struct blkcg_policy *pol = blkcg_policy[i];
+		struct blkcg_policy_data *cpd;
+
+		/*
+		 * If the policy hasn't been attached yet, wait for it
+		 * to be attached before doing anything else. Otherwise,
+		 * check if the policy requires any specific per-cgroup
+		 * data: if it does, allocate and initialize it.
+		 */
+		if (!pol || !pol->cpd_size)
+			continue;
+
+		BUG_ON(blkcg->pd[i]);
+		cpd = kzalloc(pol->cpd_size, GFP_KERNEL);
+		if (!cpd) {
+			ret = ERR_PTR(-ENOMEM);
+			goto free_pd_blkcg;
+		}
+		blkcg->pd[i] = cpd;
+		cpd->plid = i;
+		pol->cpd_init_fn(blkcg);
+	}
 
-	blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT;
-	blkcg->cfq_leaf_weight = CFQ_WEIGHT_DEFAULT;
 done:
 	spin_lock_init(&blkcg->lock);
 	INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC);
 	INIT_HLIST_HEAD(&blkcg->blkg_list);
 
 	return &blkcg->css;
+
+free_pd_blkcg:
+	for (i--; i >= 0; i--)
+		kfree(blkcg->pd[i]);
+
+free_blkcg:
+	kfree(blkcg);
+	return ret;
 }
 
 /**
@@ -958,8 +995,10 @@ int blkcg_activate_policy(struct request_queue *q,
 			  const struct blkcg_policy *pol)
 {
 	LIST_HEAD(pds);
+	LIST_HEAD(cpds);
 	struct blkcg_gq *blkg, *new_blkg;
-	struct blkg_policy_data *pd, *n;
+	struct blkg_policy_data *pd, *nd;
+	struct blkcg_policy_data *cpd, *cnd;
 	int cnt = 0, ret;
 	bool preloaded;
 
@@ -1003,7 +1042,10 @@ int blkcg_activate_policy(struct request_queue *q,
 
 	spin_unlock_irq(q->queue_lock);
 
-	/* allocate policy_data for all existing blkgs */
+	/*
+	 * Allocate per-blkg and per-blkcg policy data
+	 * for all existing blkgs.
+	 */
 	while (cnt--) {
 		pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node);
 		if (!pd) {
@@ -1011,26 +1053,50 @@ int blkcg_activate_policy(struct request_queue *q,
 			goto out_free;
 		}
 		list_add_tail(&pd->alloc_node, &pds);
+
+		if (!pol->cpd_size)
+			continue;
+		cpd = kzalloc_node(pol->cpd_size, GFP_KERNEL, q->node);
+		if (!cpd) {
+			ret = -ENOMEM;
+			goto out_free;
+		}
+		list_add_tail(&cpd->alloc_node, &cpds);
 	}
 
 	/*
-	 * Install the allocated pds.  With @q bypassing, no new blkg
+	 * Install the allocated pds and cpds. With @q bypassing, no new blkg
 	 * should have been created while the queue lock was dropped.
 	 */
 	spin_lock_irq(q->queue_lock);
 
 	list_for_each_entry(blkg, &q->blkg_list, q_node) {
-		if (WARN_ON(list_empty(&pds))) {
+		if (WARN_ON(list_empty(&pds)) ||
+		    WARN_ON(pol->cpd_size && list_empty(&cpds))) {
 			/* umm... this shouldn't happen, just abort */
 			ret = -ENOMEM;
 			goto out_unlock;
 		}
+		cpd = list_first_entry(&cpds, struct blkcg_policy_data,
+				       alloc_node);
+		list_del_init(&cpd->alloc_node);
 		pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node);
 		list_del_init(&pd->alloc_node);
 
 		/* grab blkcg lock too while installing @pd on @blkg */
 		spin_lock(&blkg->blkcg->lock);
 
+		if (!pol->cpd_size)
+			goto no_cpd;
+		if (!blkg->blkcg->pd[pol->plid]) {
+			/* Per-policy per-blkcg data */
+			blkg->blkcg->pd[pol->plid] = cpd;
+			cpd->plid = pol->plid;
+			pol->cpd_init_fn(blkg->blkcg);
+		} else { /* must free it as it has already been extracted */
+			kfree(cpd);
+		}
+no_cpd:
 		blkg->pd[pol->plid] = pd;
 		pd->blkg = blkg;
 		pd->plid = pol->plid;
@@ -1045,8 +1111,10 @@ out_unlock:
 	spin_unlock_irq(q->queue_lock);
 out_free:
 	blk_queue_bypass_end(q);
-	list_for_each_entry_safe(pd, n, &pds, alloc_node)
+	list_for_each_entry_safe(pd, nd, &pds, alloc_node)
 		kfree(pd);
+	list_for_each_entry_safe(cpd, cnd, &cpds, alloc_node)
+		kfree(cpd);
 	return ret;
 }
 EXPORT_SYMBOL_GPL(blkcg_activate_policy);
@@ -1087,6 +1155,8 @@ void blkcg_deactivate_policy(struct request_queue *q,
 
 		kfree(blkg->pd[pol->plid]);
 		blkg->pd[pol->plid] = NULL;
+		kfree(blkg->blkcg->pd[pol->plid]);
+		blkg->blkcg->pd[pol->plid] = NULL;
 
 		spin_unlock(&blkg->blkcg->lock);
 	}
diff --git a/block/blk-cgroup.h b/block/blk-cgroup.h
index c567865..74296a7 100644
--- a/block/blk-cgroup.h
+++ b/block/blk-cgroup.h
@@ -23,11 +23,6 @@
 /* Max limits for throttle policy */
 #define THROTL_IOPS_MAX		UINT_MAX
 
-/* CFQ specific, out here for blkcg->cfq_weight */
-#define CFQ_WEIGHT_MIN		10
-#define CFQ_WEIGHT_MAX		1000
-#define CFQ_WEIGHT_DEFAULT	500
-
 #ifdef CONFIG_BLK_CGROUP
 
 enum blkg_rwstat_type {
@@ -50,9 +45,7 @@ struct blkcg {
 	struct blkcg_gq			*blkg_hint;
 	struct hlist_head		blkg_list;
 
-	/* TODO: per-policy storage in blkcg */
-	unsigned int			cfq_weight;	/* belongs to cfq */
-	unsigned int			cfq_leaf_weight;
+	struct blkcg_policy_data	*pd[BLKCG_MAX_POLS];
 };
 
 struct blkg_stat {
@@ -87,6 +80,24 @@ struct blkg_policy_data {
 	struct list_head		alloc_node;
 };
 
+/*
+ * Policies that need to keep per-blkcg data which is independent
+ * from any request_queue associated to it must specify its size
+ * with the cpd_size field of the blkcg_policy structure and
+ * embed a blkcg_policy_data in it. blkcg core allocates
+ * policy-specific per-blkcg structures lazily the first time
+ * they are actually needed, so it handles them together with
+ * blkgs. cpd_init() is invoked to let each policy handle
+ * per-blkcg data.
+ */
+struct blkcg_policy_data {
+	/* the policy id this per-policy data belongs to */
+	int				plid;
+
+	/* used during policy activation */
+	struct list_head		alloc_node;
+};
+
 /* association between a blk cgroup and a request queue */
 struct blkcg_gq {
 	/* Pointer to the associated request_queue */
@@ -112,6 +123,7 @@ struct blkcg_gq {
 	struct rcu_head			rcu_head;
 };
 
+typedef void (blkcg_pol_init_cpd_fn)(const struct blkcg *blkcg);
 typedef void (blkcg_pol_init_pd_fn)(struct blkcg_gq *blkg);
 typedef void (blkcg_pol_online_pd_fn)(struct blkcg_gq *blkg);
 typedef void (blkcg_pol_offline_pd_fn)(struct blkcg_gq *blkg);
@@ -122,10 +134,13 @@ struct blkcg_policy {
 	int				plid;
 	/* policy specific private data size */
 	size_t				pd_size;
+	/* policy specific per-blkcg data size */
+	size_t				cpd_size;
 	/* cgroup files for the policy */
 	struct cftype			*cftypes;
 
 	/* operations */
+	blkcg_pol_init_cpd_fn		*cpd_init_fn;
 	blkcg_pol_init_pd_fn		*pd_init_fn;
 	blkcg_pol_online_pd_fn		*pd_online_fn;
 	blkcg_pol_offline_pd_fn		*pd_offline_fn;
@@ -218,6 +233,12 @@ static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
 	return blkg ? blkg->pd[pol->plid] : NULL;
 }
 
+static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
+						     struct blkcg_policy *pol)
+{
+	return blkcg ? blkcg->pd[pol->plid] : NULL;
+}
+
 /**
  * pdata_to_blkg - get blkg associated with policy private data
  * @pd: policy private data of interest
@@ -564,6 +585,9 @@ struct blkcg;
 struct blkg_policy_data {
 };
 
+struct blkcg_policy_data {
+};
+
 struct blkcg_gq {
 };
 
diff --git a/block/cfq-iosched.c b/block/cfq-iosched.c
index 402be01..c808ad8 100644
--- a/block/cfq-iosched.c
+++ b/block/cfq-iosched.c
@@ -67,6 +67,11 @@ static struct kmem_cache *cfq_pool;
 #define sample_valid(samples)	((samples) > 80)
 #define rb_entry_cfqg(node)	rb_entry((node), struct cfq_group, rb_node)
 
+/* blkio-related constants */
+#define CFQ_WEIGHT_MIN          10
+#define CFQ_WEIGHT_MAX          1000
+#define CFQ_WEIGHT_DEFAULT      500
+
 struct cfq_ttime {
 	unsigned long last_end_request;
 
@@ -212,6 +217,15 @@ struct cfqg_stats {
 #endif	/* CONFIG_CFQ_GROUP_IOSCHED */
 };
 
+/* Per-cgroup data */
+struct cfq_group_data {
+	/* must be the first member */
+	struct blkcg_policy_data pd;
+
+	unsigned int weight;
+	unsigned int leaf_weight;
+};
+
 /* This is per cgroup per device grouping structure */
 struct cfq_group {
 	/* must be the first member */
@@ -451,6 +465,12 @@ static inline struct cfq_group *pd_to_cfqg(struct blkg_policy_data *pd)
 	return pd ? container_of(pd, struct cfq_group, pd) : NULL;
 }
 
+static struct cfq_group_data
+*cpd_to_cfqgd(struct blkcg_policy_data *cpd)
+{
+	return cpd ? container_of(cpd, struct cfq_group_data, pd) : NULL;
+}
+
 static inline struct blkcg_gq *cfqg_to_blkg(struct cfq_group *cfqg)
 {
 	return pd_to_blkg(&cfqg->pd);
@@ -607,6 +627,11 @@ static inline struct cfq_group *blkg_to_cfqg(struct blkcg_gq *blkg)
 	return pd_to_cfqg(blkg_to_pd(blkg, &blkcg_policy_cfq));
 }
 
+static struct cfq_group_data *blkcg_to_cfqgd(struct blkcg *blkcg)
+{
+	return cpd_to_cfqgd(blkcg_to_cpd(blkcg, &blkcg_policy_cfq));
+}
+
 static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg)
 {
 	struct blkcg_gq *pblkg = cfqg_to_blkg(cfqg)->parent;
@@ -1544,13 +1569,28 @@ static void cfqg_stats_init(struct cfqg_stats *stats)
 #endif
 }
 
+static void cfq_cpd_init(const struct blkcg *blkcg)
+{
+	struct cfq_group_data *cgd =
+		cpd_to_cfqgd(blkcg->pd[blkcg_policy_cfq.plid]);
+
+	if (blkcg == &blkcg_root) {
+		cgd->weight = 2 * CFQ_WEIGHT_DEFAULT;
+		cgd->leaf_weight = 2 * CFQ_WEIGHT_DEFAULT;
+	} else {
+		cgd->weight = CFQ_WEIGHT_DEFAULT;
+		cgd->leaf_weight = CFQ_WEIGHT_DEFAULT;
+	}
+}
+
 static void cfq_pd_init(struct blkcg_gq *blkg)
 {
 	struct cfq_group *cfqg = blkg_to_cfqg(blkg);
+	struct cfq_group_data *cgd = blkcg_to_cfqgd(blkg->blkcg);
 
 	cfq_init_cfqg_base(cfqg);
-	cfqg->weight = blkg->blkcg->cfq_weight;
-	cfqg->leaf_weight = blkg->blkcg->cfq_leaf_weight;
+	cfqg->weight = cgd->weight;
+	cfqg->leaf_weight = cgd->leaf_weight;
 	cfqg_stats_init(&cfqg->stats);
 	cfqg_stats_init(&cfqg->dead_stats);
 }
@@ -1673,13 +1713,17 @@ static int cfqg_print_leaf_weight_device(struct seq_file *sf, void *v)
 
 static int cfq_print_weight(struct seq_file *sf, void *v)
 {
-	seq_printf(sf, "%u\n", css_to_blkcg(seq_css(sf))->cfq_weight);
+	struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
+
+	seq_printf(sf, "%u\n", blkcg_to_cfqgd(blkcg)->weight);
 	return 0;
 }
 
 static int cfq_print_leaf_weight(struct seq_file *sf, void *v)
 {
-	seq_printf(sf, "%u\n", css_to_blkcg(seq_css(sf))->cfq_leaf_weight);
+	struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
+
+	seq_printf(sf, "%u\n", blkcg_to_cfqgd(blkcg)->leaf_weight);
 	return 0;
 }
 
@@ -1690,6 +1734,7 @@ static ssize_t __cfqg_set_weight_device(struct kernfs_open_file *of,
 	struct blkcg *blkcg = css_to_blkcg(of_css(of));
 	struct blkg_conf_ctx ctx;
 	struct cfq_group *cfqg;
+	struct cfq_group_data *cfqgd;
 	int ret;
 
 	ret = blkg_conf_prep(blkcg, &blkcg_policy_cfq, buf, &ctx);
@@ -1698,13 +1743,14 @@ static ssize_t __cfqg_set_weight_device(struct kernfs_open_file *of,
 
 	ret = -EINVAL;
 	cfqg = blkg_to_cfqg(ctx.blkg);
+	cfqgd = blkcg_to_cfqgd(blkcg);
 	if (!ctx.v || (ctx.v >= CFQ_WEIGHT_MIN && ctx.v <= CFQ_WEIGHT_MAX)) {
 		if (!is_leaf_weight) {
 			cfqg->dev_weight = ctx.v;
-			cfqg->new_weight = ctx.v ?: blkcg->cfq_weight;
+			cfqg->new_weight = ctx.v ?: cfqgd->weight;
 		} else {
 			cfqg->dev_leaf_weight = ctx.v;
-			cfqg->new_leaf_weight = ctx.v ?: blkcg->cfq_leaf_weight;
+			cfqg->new_leaf_weight = ctx.v ?: cfqgd->leaf_weight;
 		}
 		ret = 0;
 	}
@@ -1730,16 +1776,18 @@ static int __cfq_set_weight(struct cgroup_subsys_state *css, struct cftype *cft,
 {
 	struct blkcg *blkcg = css_to_blkcg(css);
 	struct blkcg_gq *blkg;
+	struct cfq_group_data *cfqgd;
 
 	if (val < CFQ_WEIGHT_MIN || val > CFQ_WEIGHT_MAX)
 		return -EINVAL;
 
 	spin_lock_irq(&blkcg->lock);
+	cfqgd = blkcg_to_cfqgd(blkcg);
 
 	if (!is_leaf_weight)
-		blkcg->cfq_weight = val;
+		cfqgd->weight = val;
 	else
-		blkcg->cfq_leaf_weight = val;
+		cfqgd->leaf_weight = val;
 
 	hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
 		struct cfq_group *cfqg = blkg_to_cfqg(blkg);
@@ -1749,10 +1797,10 @@ static int __cfq_set_weight(struct cgroup_subsys_state *css, struct cftype *cft,
 
 		if (!is_leaf_weight) {
 			if (!cfqg->dev_weight)
-				cfqg->new_weight = blkcg->cfq_weight;
+				cfqg->new_weight = cfqgd->weight;
 		} else {
 			if (!cfqg->dev_leaf_weight)
-				cfqg->new_leaf_weight = blkcg->cfq_leaf_weight;
+				cfqg->new_leaf_weight = cfqgd->leaf_weight;
 		}
 	}
 
@@ -4603,8 +4651,10 @@ static struct elevator_type iosched_cfq = {
 #ifdef CONFIG_CFQ_GROUP_IOSCHED
 static struct blkcg_policy blkcg_policy_cfq = {
 	.pd_size		= sizeof(struct cfq_group),
+	.cpd_size		= sizeof(struct cfq_group_data),
 	.cftypes		= cfq_blkcg_files,
 
+	.cpd_init_fn		= cfq_cpd_init,
 	.pd_init_fn		= cfq_pd_init,
 	.pd_offline_fn		= cfq_pd_offline,
 	.pd_reset_stats_fn	= cfq_pd_reset_stats,