Merge branch 'blkcg-cfq-hierarchy' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup into for-3.9/core

Tejun writes:

Hello, Jens.

Please consider pulling from the following branch to receive cfq blkcg
hierarchy support.  The branch is based on top of v3.8-rc2.

  git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup.git blkcg-cfq-hierarchy

The patchset was reviewd in the following thread.

  http://thread.gmane.org/gmane.linux.kernel.cgroups/5571

7 files changed, 902 insertions, 174 deletions

diff --git a/Documentation/block/cfq-iosched.txt b/Documentation/block/cfq-iosched.txt
index d89b4fe724d..a5eb7d19a65 100644
--- a/Documentation/block/cfq-iosched.txt
+++ b/Documentation/block/cfq-iosched.txt
@@ -102,6 +102,64 @@ processing of request. Therefore, increasing the value can imporve the
 performace although this can cause the latency of some I/O to increase due
 to more number of requests.
 
+CFQ Group scheduling
+====================
+
+CFQ supports blkio cgroup and has "blkio." prefixed files in each
+blkio cgroup directory. It is weight-based and there are four knobs
+for configuration - weight[_device] and leaf_weight[_device].
+Internal cgroup nodes (the ones with children) can also have tasks in
+them, so the former two configure how much proportion the cgroup as a
+whole is entitled to at its parent's level while the latter two
+configure how much proportion the tasks in the cgroup have compared to
+its direct children.
+
+Another way to think about it is assuming that each internal node has
+an implicit leaf child node which hosts all the tasks whose weight is
+configured by leaf_weight[_device]. Let's assume a blkio hierarchy
+composed of five cgroups - root, A, B, AA and AB - with the following
+weights where the names represent the hierarchy.
+
+        weight leaf_weight
+ root :  125    125
+ A    :  500    750
+ B    :  250    500
+ AA   :  500    500
+ AB   : 1000    500
+
+root never has a parent making its weight is meaningless. For backward
+compatibility, weight is always kept in sync with leaf_weight. B, AA
+and AB have no child and thus its tasks have no children cgroup to
+compete with. They always get 100% of what the cgroup won at the
+parent level. Considering only the weights which matter, the hierarchy
+looks like the following.
+
+          root
+       /    |   \
+      A     B    leaf
+     500   250   125
+   /  |  \
+  AA  AB  leaf
+ 500 1000 750
+
+If all cgroups have active IOs and competing with each other, disk
+time will be distributed like the following.
+
+Distribution below root. The total active weight at this level is
+A:500 + B:250 + C:125 = 875.
+
+ root-leaf :   125 /  875      =~ 14%
+ A         :   500 /  875      =~ 57%
+ B(-leaf)  :   250 /  875      =~ 28%
+
+A has children and further distributes its 57% among the children and
+the implicit leaf node. The total active weight at this level is
+AA:500 + AB:1000 + A-leaf:750 = 2250.
+
+ A-leaf    : ( 750 / 2250) * A =~ 19%
+ AA(-leaf) : ( 500 / 2250) * A =~ 12%
+ AB(-leaf) : (1000 / 2250) * A =~ 25%
+
 CFQ IOPS Mode for group scheduling
 ===================================
 Basic CFQ design is to provide priority based time slices. Higher priority
diff --git a/Documentation/cgroups/blkio-controller.txt b/Documentation/cgroups/blkio-controller.txt
index b4b1fb3a83f..1b70843c574 100644
--- a/Documentation/cgroups/blkio-controller.txt
+++ b/Documentation/cgroups/blkio-controller.txt
@@ -94,13 +94,11 @@ Throttling/Upper Limit policy
 
 Hierarchical Cgroups
 ====================
-- Currently none of the IO control policy supports hierarchical groups. But
-  cgroup interface does allow creation of hierarchical cgroups and internally
-  IO policies treat them as flat hierarchy.
+- Currently only CFQ supports hierarchical groups. For throttling,
+  cgroup interface does allow creation of hierarchical cgroups and
+  internally it treats them as flat hierarchy.
 
-  So this patch will allow creation of cgroup hierarchcy but at the backend
-  everything will be treated as flat. So if somebody created a hierarchy like
-  as follows.
+  If somebody created a hierarchy like as follows.
 
 			root
 			/  \
@@ -108,16 +106,20 @@ Hierarchical Cgroups
 			|
 		     test3
 
-  CFQ and throttling will practically treat all groups at same level.
+  CFQ will handle the hierarchy correctly but and throttling will
+  practically treat all groups at same level. For details on CFQ
+  hierarchy support, refer to Documentation/block/cfq-iosched.txt.
+  Throttling will treat the hierarchy as if it looks like the
+  following.
 
 				pivot
 			     /  /   \  \
 			root  test1 test2  test3
 
-  Down the line we can implement hierarchical accounting/control support
-  and also introduce a new cgroup file "use_hierarchy" which will control
-  whether cgroup hierarchy is viewed as flat or hierarchical by the policy..
-  This is how memory controller also has implemented the things.
+  Nesting cgroups, while allowed, isn't officially supported and blkio
+  genereates warning when cgroups nest. Once throttling implements
+  hierarchy support, hierarchy will be supported and the warning will
+  be removed.
 
 Various user visible config options
 ===================================
@@ -172,6 +174,12 @@ Proportional weight policy files
 	  dev     weight
 	  8:16    300
 
+- blkio.leaf_weight[_device]
+	- Equivalents of blkio.weight[_device] for the purpose of
+          deciding how much weight tasks in the given cgroup has while
+          competing with the cgroup's child cgroups. For details,
+          please refer to Documentation/block/cfq-iosched.txt.
+
 - blkio.time
 	- disk time allocated to cgroup per device in milliseconds. First
 	  two fields specify the major and minor number of the device and
@@ -279,6 +287,11 @@ Proportional weight policy files
 	  and minor number of the device and third field specifies the number
 	  of times a group was dequeued from a particular device.
 
+- blkio.*_recursive
+	- Recursive version of various stats. These files show the
+          same information as their non-recursive counterparts but
+          include stats from all the descendant cgroups.
+
 Throttling/Upper limit policy files
 -----------------------------------
 - blkio.throttle.read_bps_device
diff --git a/block/blk-cgroup.c b/block/blk-cgroup.c
index b8858fb0caf..87ea95d1f53 100644
--- a/block/blk-cgroup.c
+++ b/block/blk-cgroup.c
@@ -26,11 +26,32 @@
 
 static DEFINE_MUTEX(blkcg_pol_mutex);
 
-struct blkcg blkcg_root = { .cfq_weight = 2 * CFQ_WEIGHT_DEFAULT };
+struct blkcg blkcg_root = { .cfq_weight = 2 * CFQ_WEIGHT_DEFAULT,
+			    .cfq_leaf_weight = 2 * CFQ_WEIGHT_DEFAULT, };
 EXPORT_SYMBOL_GPL(blkcg_root);
 
 static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
 
+static struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
+				      struct request_queue *q, bool update_hint);
+
+/**
+ * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
+ * @d_blkg: loop cursor pointing to the current descendant
+ * @pos_cgrp: used for iteration
+ * @p_blkg: target blkg to walk descendants of
+ *
+ * Walk @c_blkg through the descendants of @p_blkg.  Must be used with RCU
+ * read locked.  If called under either blkcg or queue lock, the iteration
+ * is guaranteed to include all and only online blkgs.  The caller may
+ * update @pos_cgrp by calling cgroup_rightmost_descendant() to skip
+ * subtree.
+ */
+#define blkg_for_each_descendant_pre(d_blkg, pos_cgrp, p_blkg)		\
+	cgroup_for_each_descendant_pre((pos_cgrp), (p_blkg)->blkcg->css.cgroup) \
+		if (((d_blkg) = __blkg_lookup(cgroup_to_blkcg(pos_cgrp), \
+					      (p_blkg)->q, false)))
+
 static bool blkcg_policy_enabled(struct request_queue *q,
 				 const struct blkcg_policy *pol)
 {
@@ -112,9 +133,10 @@ static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
 
 		blkg->pd[i] = pd;
 		pd->blkg = blkg;
+		pd->plid = i;
 
 		/* invoke per-policy init */
-		if (blkcg_policy_enabled(blkg->q, pol))
+		if (pol->pd_init_fn)
 			pol->pd_init_fn(blkg);
 	}
 
@@ -125,8 +147,19 @@ err_free:
 	return NULL;
 }
 
+/**
+ * __blkg_lookup - internal version of blkg_lookup()
+ * @blkcg: blkcg of interest
+ * @q: request_queue of interest
+ * @update_hint: whether to update lookup hint with the result or not
+ *
+ * This is internal version and shouldn't be used by policy
+ * implementations.  Looks up blkgs for the @blkcg - @q pair regardless of
+ * @q's bypass state.  If @update_hint is %true, the caller should be
+ * holding @q->queue_lock and lookup hint is updated on success.
+ */
 static struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
-				      struct request_queue *q)
+				      struct request_queue *q, bool update_hint)
 {
 	struct blkcg_gq *blkg;
 
@@ -135,14 +168,19 @@ static struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
 		return blkg;
 
 	/*
-	 * Hint didn't match.  Look up from the radix tree.  Note that we
-	 * may not be holding queue_lock and thus are not sure whether
-	 * @blkg from blkg_tree has already been removed or not, so we
-	 * can't update hint to the lookup result.  Leave it to the caller.
+	 * Hint didn't match.  Look up from the radix tree.  Note that the
+	 * hint can only be updated under queue_lock as otherwise @blkg
+	 * could have already been removed from blkg_tree.  The caller is
+	 * responsible for grabbing queue_lock if @update_hint.
 	 */
 	blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
-	if (blkg && blkg->q == q)
+	if (blkg && blkg->q == q) {
+		if (update_hint) {
+			lockdep_assert_held(q->queue_lock);
+			rcu_assign_pointer(blkcg->blkg_hint, blkg);
+		}
 		return blkg;
+	}
 
 	return NULL;
 }
@@ -162,7 +200,7 @@ struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q)
 
 	if (unlikely(blk_queue_bypass(q)))
 		return NULL;
-	return __blkg_lookup(blkcg, q);
+	return __blkg_lookup(blkcg, q, false);
 }
 EXPORT_SYMBOL_GPL(blkg_lookup);
 
@@ -170,75 +208,129 @@ EXPORT_SYMBOL_GPL(blkg_lookup);
  * If @new_blkg is %NULL, this function tries to allocate a new one as
  * necessary using %GFP_ATOMIC.  @new_blkg is always consumed on return.
  */
-static struct blkcg_gq *__blkg_lookup_create(struct blkcg *blkcg,
-					     struct request_queue *q,
-					     struct blkcg_gq *new_blkg)
+static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
+				    struct request_queue *q,
+				    struct blkcg_gq *new_blkg)
 {
 	struct blkcg_gq *blkg;
-	int ret;
+	int i, ret;
 
 	WARN_ON_ONCE(!rcu_read_lock_held());
 	lockdep_assert_held(q->queue_lock);
 
-	/* lookup and update hint on success, see __blkg_lookup() for details */
-	blkg = __blkg_lookup(blkcg, q);
-	if (blkg) {
-		rcu_assign_pointer(blkcg->blkg_hint, blkg);
-		goto out_free;
-	}
-
 	/* blkg holds a reference to blkcg */
 	if (!css_tryget(&blkcg->css)) {
-		blkg = ERR_PTR(-EINVAL);
-		goto out_free;
+		ret = -EINVAL;
+		goto err_free_blkg;
 	}
 
 	/* allocate */
 	if (!new_blkg) {
 		new_blkg = blkg_alloc(blkcg, q, GFP_ATOMIC);
 		if (unlikely(!new_blkg)) {
-			blkg = ERR_PTR(-ENOMEM);
-			goto out_put;
+			ret = -ENOMEM;
+			goto err_put_css;
 		}
 	}
 	blkg = new_blkg;
 
-	/* insert */
+	/* link parent and insert */
+	if (blkcg_parent(blkcg)) {
+		blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false);
+		if (WARN_ON_ONCE(!blkg->parent)) {
+			blkg = ERR_PTR(-EINVAL);
+			goto err_put_css;
+		}
+		blkg_get(blkg->parent);
+	}
+
 	spin_lock(&blkcg->lock);
 	ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
 	if (likely(!ret)) {
 		hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
 		list_add(&blkg->q_node, &q->blkg_list);
+
+		for (i = 0; i < BLKCG_MAX_POLS; i++) {
+			struct blkcg_policy *pol = blkcg_policy[i];
+
+			if (blkg->pd[i] && pol->pd_online_fn)
+				pol->pd_online_fn(blkg);
+		}
 	}
+	blkg->online = true;
 	spin_unlock(&blkcg->lock);
 
 	if (!ret)
 		return blkg;
 
-	blkg = ERR_PTR(ret);
-out_put:
+	/* @blkg failed fully initialized, use the usual release path */
+	blkg_put(blkg);
+	return ERR_PTR(ret);
+
+err_put_css:
 	css_put(&blkcg->css);
-out_free:
+err_free_blkg:
 	blkg_free(new_blkg);
-	return blkg;
+	return ERR_PTR(ret);
 }
 
+/**
+ * blkg_lookup_create - lookup blkg, try to create one if not there
+ * @blkcg: blkcg of interest
+ * @q: request_queue of interest
+ *
+ * Lookup blkg for the @blkcg - @q pair.  If it doesn't exist, try to
+ * create one.  blkg creation is performed recursively from blkcg_root such
+ * that all non-root blkg's have access to the parent blkg.  This function
+ * should be called under RCU read lock and @q->queue_lock.
+ *
+ * Returns pointer to the looked up or created blkg on success, ERR_PTR()
+ * value on error.  If @q is dead, returns ERR_PTR(-EINVAL).  If @q is not
+ * dead and bypassing, returns ERR_PTR(-EBUSY).
+ */
 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
 				    struct request_queue *q)
 {
+	struct blkcg_gq *blkg;
+
+	WARN_ON_ONCE(!rcu_read_lock_held());
+	lockdep_assert_held(q->queue_lock);
+
 	/*
 	 * This could be the first entry point of blkcg implementation and
 	 * we shouldn't allow anything to go through for a bypassing queue.
 	 */
 	if (unlikely(blk_queue_bypass(q)))
 		return ERR_PTR(blk_queue_dying(q) ? -EINVAL : -EBUSY);
-	return __blkg_lookup_create(blkcg, q, NULL);
+
+	blkg = __blkg_lookup(blkcg, q, true);
+	if (blkg)
+		return blkg;
+
+	/*
+	 * Create blkgs walking down from blkcg_root to @blkcg, so that all
+	 * non-root blkgs have access to their parents.
+	 */
+	while (true) {
+		struct blkcg *pos = blkcg;
+		struct blkcg *parent = blkcg_parent(blkcg);
+
+		while (parent && !__blkg_lookup(parent, q, false)) {
+			pos = parent;
+			parent = blkcg_parent(parent);
+		}
+
+		blkg = blkg_create(pos, q, NULL);
+		if (pos == blkcg || IS_ERR(blkg))
+			return blkg;
+	}
 }
 EXPORT_SYMBOL_GPL(blkg_lookup_create);
 
 static void blkg_destroy(struct blkcg_gq *blkg)
 {
 	struct blkcg *blkcg = blkg->blkcg;
+	int i;
 
 	lockdep_assert_held(blkg->q->queue_lock);
 	lockdep_assert_held(&blkcg->lock);
@@ -247,6 +339,14 @@ static void blkg_destroy(struct blkcg_gq *blkg)
 	WARN_ON_ONCE(list_empty(&blkg->q_node));
 	WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
 
+	for (i = 0; i < BLKCG_MAX_POLS; i++) {
+		struct blkcg_policy *pol = blkcg_policy[i];
+
+		if (blkg->pd[i] && pol->pd_offline_fn)
+			pol->pd_offline_fn(blkg);
+	}
+	blkg->online = false;
+
 	radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
 	list_del_init(&blkg->q_node);
 	hlist_del_init_rcu(&blkg->blkcg_node);
@@ -301,8 +401,10 @@ static void blkg_rcu_free(struct rcu_head *rcu_head)
 
 void __blkg_release(struct blkcg_gq *blkg)
 {
-	/* release the extra blkcg reference this blkg has been holding */
+	/* release the blkcg and parent blkg refs this blkg has been holding */
 	css_put(&blkg->blkcg->css);
+	if (blkg->parent)
+		blkg_put(blkg->parent);
 
 	/*
 	 * A group is freed in rcu manner. But having an rcu lock does not
@@ -402,8 +504,9 @@ static const char *blkg_dev_name(struct blkcg_gq *blkg)
  *
  * This function invokes @prfill on each blkg of @blkcg if pd for the
  * policy specified by @pol exists.  @prfill is invoked with @sf, the
- * policy data and @data.  If @show_total is %true, the sum of the return
- * values from @prfill is printed with "Total" label at the end.
+ * policy data and @data and the matching queue lock held.  If @show_total
+ * is %true, the sum of the return values from @prfill is printed with
+ * "Total" label at the end.
  *
  * This is to be used to construct print functions for
  * cftype->read_seq_string method.
@@ -418,11 +521,14 @@ void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
 	struct hlist_node *n;
 	u64 total = 0;
 
-	spin_lock_irq(&blkcg->lock);
-	hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node)
+	rcu_read_lock();
+	hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
+		spin_lock_irq(blkg->q->queue_lock);
 		if (blkcg_policy_enabled(blkg->q, pol))
 			total += prfill(sf, blkg->pd[pol->plid], data);
-	spin_unlock_irq(&blkcg->lock);
+		spin_unlock_irq(blkg->q->queue_lock);
+	}
+	rcu_read_unlock();
 
 	if (show_total)
 		seq_printf(sf, "Total %llu\n", (unsigned long long)total);
@@ -481,6 +587,7 @@ u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
 	seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
 	return v;
 }
+EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
 
 /**
  * blkg_prfill_stat - prfill callback for blkg_stat
@@ -514,6 +621,82 @@ u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
 EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
 
 /**
+ * blkg_stat_recursive_sum - collect hierarchical blkg_stat
+ * @pd: policy private data of interest
+ * @off: offset to the blkg_stat in @pd
+ *
+ * Collect the blkg_stat specified by @off from @pd and all its online
+ * descendants and return the sum.  The caller must be holding the queue
+ * lock for online tests.
+ */
+u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off)
+{
+	struct blkcg_policy *pol = blkcg_policy[pd->plid];
+	struct blkcg_gq *pos_blkg;
+	struct cgroup *pos_cgrp;
+	u64 sum;
+
+	lockdep_assert_held(pd->blkg->q->queue_lock);
+
+	sum = blkg_stat_read((void *)pd + off);
+
+	rcu_read_lock();
+	blkg_for_each_descendant_pre(pos_blkg, pos_cgrp, pd_to_blkg(pd)) {
+		struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
+		struct blkg_stat *stat = (void *)pos_pd + off;
+
+		if (pos_blkg->online)
+			sum += blkg_stat_read(stat);
+	}
+	rcu_read_unlock();
+
+	return sum;
+}
+EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum);
+
+/**
+ * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
+ * @pd: policy private data of interest
+ * @off: offset to the blkg_stat in @pd
+ *
+ * Collect the blkg_rwstat specified by @off from @pd and all its online
+ * descendants and return the sum.  The caller must be holding the queue
+ * lock for online tests.
+ */
+struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
+					     int off)
+{
+	struct blkcg_policy *pol = blkcg_policy[pd->plid];
+	struct blkcg_gq *pos_blkg;
+	struct cgroup *pos_cgrp;
+	struct blkg_rwstat sum;
+	int i;
+
+	lockdep_assert_held(pd->blkg->q->queue_lock);
+
+	sum = blkg_rwstat_read((void *)pd + off);
+
+	rcu_read_lock();
+	blkg_for_each_descendant_pre(pos_blkg, pos_cgrp, pd_to_blkg(pd)) {
+		struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
+		struct blkg_rwstat *rwstat = (void *)pos_pd + off;
+		struct blkg_rwstat tmp;
+
+		if (!pos_blkg->online)
+			continue;
+
+		tmp = blkg_rwstat_read(rwstat);
+
+		for (i = 0; i < BLKG_RWSTAT_NR; i++)
+			sum.cnt[i] += tmp.cnt[i];
+	}
+	rcu_read_unlock();
+
+	return sum;
+}
+EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
+
+/**
  * blkg_conf_prep - parse and prepare for per-blkg config update
  * @blkcg: target block cgroup
  * @pol: target policy
@@ -658,6 +841,7 @@ static struct cgroup_subsys_state *blkcg_css_alloc(struct cgroup *cgroup)
 		return ERR_PTR(-ENOMEM);
 
 	blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT;
+	blkcg->cfq_leaf_weight = CFQ_WEIGHT_DEFAULT;
 	blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */
 done:
 	spin_lock_init(&blkcg->lock);
@@ -777,7 +961,7 @@ int blkcg_activate_policy(struct request_queue *q,
 			  const struct blkcg_policy *pol)
 {
 	LIST_HEAD(pds);
-	struct blkcg_gq *blkg;
+	struct blkcg_gq *blkg, *new_blkg;
 	struct blkg_policy_data *pd, *n;
 	int cnt = 0, ret;
 	bool preloaded;
@@ -786,19 +970,27 @@ int blkcg_activate_policy(struct request_queue *q,
 		return 0;
 
 	/* preallocations for root blkg */
-	blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
-	if (!blkg)
+	new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
+	if (!new_blkg)
 		return -ENOMEM;
 
 	preloaded = !radix_tree_preload(GFP_KERNEL);
 
 	blk_queue_bypass_start(q);
 
-	/* make sure the root blkg exists and count the existing blkgs */
+	/*
+	 * Make sure the root blkg exists and count the existing blkgs.  As
+	 * @q is bypassing at this point, blkg_lookup_create() can't be
+	 * used.  Open code it.
+	 */
 	spin_lock_irq(q->queue_lock);
 
 	rcu_read_lock();
-	blkg = __blkg_lookup_create(&blkcg_root, q, blkg);
+	blkg = __blkg_lookup(&blkcg_root, q, false);
+	if (blkg)
+		blkg_free(new_blkg);
+	else
+		blkg = blkg_create(&blkcg_root, q, new_blkg);
 	rcu_read_unlock();
 
 	if (preloaded)
@@ -846,6 +1038,7 @@ int blkcg_activate_policy(struct request_queue *q,
 
 		blkg->pd[pol->plid] = pd;
 		pd->blkg = blkg;
+		pd->plid = pol->plid;
 		pol->pd_init_fn(blkg);
 
 		spin_unlock(&blkg->blkcg->lock);
@@ -892,6 +1085,8 @@ void blkcg_deactivate_policy(struct request_queue *q,
 		/* grab blkcg lock too while removing @pd from @blkg */
 		spin_lock(&blkg->blkcg->lock);
 
+		if (pol->pd_offline_fn)
+			pol->pd_offline_fn(blkg);
 		if (pol->pd_exit_fn)
 			pol->pd_exit_fn(blkg);
 
diff --git a/block/blk-cgroup.h b/block/blk-cgroup.h
index 24597309e23..f2b292925cc 100644
--- a/block/blk-cgroup.h
+++ b/block/blk-cgroup.h
@@ -54,6 +54,7 @@ struct blkcg {
 
 	/* TODO: per-policy storage in blkcg */
 	unsigned int			cfq_weight;	/* belongs to cfq */
+	unsigned int			cfq_leaf_weight;
 };
 
 struct blkg_stat {
@@ -80,8 +81,9 @@ struct blkg_rwstat {
  * beginning and pd_size can't be smaller than pd.
  */
 struct blkg_policy_data {
-	/* the blkg this per-policy data belongs to */
+	/* the blkg and policy id this per-policy data belongs to */
 	struct blkcg_gq			*blkg;
+	int				plid;
 
 	/* used during policy activation */
 	struct list_head		alloc_node;
@@ -94,17 +96,27 @@ struct blkcg_gq {
 	struct list_head		q_node;
 	struct hlist_node		blkcg_node;
 	struct blkcg			*blkcg;
+
+	/* all non-root blkcg_gq's are guaranteed to have access to parent */
+	struct blkcg_gq			*parent;
+
 	/* request allocation list for this blkcg-q pair */
 	struct request_list		rl;
+
 	/* reference count */
 	int				refcnt;
 
+	/* is this blkg online? protected by both blkcg and q locks */
+	bool				online;
+
 	struct blkg_policy_data		*pd[BLKCG_MAX_POLS];
 
 	struct rcu_head			rcu_head;
 };
 
 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);
 typedef void (blkcg_pol_exit_pd_fn)(struct blkcg_gq *blkg);
 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkcg_gq *blkg);
 
@@ -117,6 +129,8 @@ struct blkcg_policy {
 
 	/* operations */
 	blkcg_pol_init_pd_fn		*pd_init_fn;
+	blkcg_pol_online_pd_fn		*pd_online_fn;
+	blkcg_pol_offline_pd_fn		*pd_offline_fn;
 	blkcg_pol_exit_pd_fn		*pd_exit_fn;
 	blkcg_pol_reset_pd_stats_fn	*pd_reset_stats_fn;
 };
@@ -150,6 +164,10 @@ u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
 		       int off);
 
+u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off);
+struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
+					     int off);
+
 struct blkg_conf_ctx {
 	struct gendisk			*disk;
 	struct blkcg_gq			*blkg;
@@ -181,6 +199,19 @@ static inline struct blkcg *bio_blkcg(struct bio *bio)
 }
 
 /**
+ * blkcg_parent - get the parent of a blkcg
+ * @blkcg: blkcg of interest
+ *
+ * Return the parent blkcg of @blkcg.  Can be called anytime.
+ */
+static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
+{
+	struct cgroup *pcg = blkcg->css.cgroup->parent;
+
+	return pcg ? cgroup_to_blkcg(pcg) : NULL;
+}
+
+/**
  * blkg_to_pdata - get policy private data
  * @blkg: blkg of interest
  * @pol: policy of interest
@@ -387,6 +418,18 @@ static inline void blkg_stat_reset(struct blkg_stat *stat)
 }
 
 /**
+ * blkg_stat_merge - merge a blkg_stat into another
+ * @to: the destination blkg_stat
+ * @from: the source
+ *
+ * Add @from's count to @to.
+ */
+static inline void blkg_stat_merge(struct blkg_stat *to, struct blkg_stat *from)
+{
+	blkg_stat_add(to, blkg_stat_read(from));
+}
+
+/**
  * blkg_rwstat_add - add a value to a blkg_rwstat
  * @rwstat: target blkg_rwstat
  * @rw: mask of REQ_{WRITE|SYNC}
@@ -434,14 +477,14 @@ static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
 }
 
 /**
- * blkg_rwstat_sum - read the total count of a blkg_rwstat
+ * blkg_rwstat_total - read the total count of a blkg_rwstat
  * @rwstat: blkg_rwstat to read
  *
  * Return the total count of @rwstat regardless of the IO direction.  This
  * function can be called without synchronization and takes care of u64
  * atomicity.
  */
-static inline uint64_t blkg_rwstat_sum(struct blkg_rwstat *rwstat)
+static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
 {
 	struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
 
@@ -457,6 +500,25 @@ static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
 	memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
 }
 
+/**
+ * blkg_rwstat_merge - merge a blkg_rwstat into another
+ * @to: the destination blkg_rwstat
+ * @from: the source
+ *
+ * Add @from's counts to @to.
+ */
+static inline void blkg_rwstat_merge(struct blkg_rwstat *to,
+				     struct blkg_rwstat *from)
+{
+	struct blkg_rwstat v = blkg_rwstat_read(from);
+	int i;
+
+	u64_stats_update_begin(&to->syncp);
+	for (i = 0; i < BLKG_RWSTAT_NR; i++)
+		to->cnt[i] += v.cnt[i];
+	u64_stats_update_end(&to->syncp);
+}
+
 #else	/* CONFIG_BLK_CGROUP */
 
 struct cgroup;
diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
index 788147797a7..6206a934eb8 100644
--- a/block/blk-sysfs.c
+++ b/block/blk-sysfs.c
@@ -497,6 +497,13 @@ queue_attr_store(struct kobject *kobj, struct attribute *attr,
 	return res;
 }
 
+static void blk_free_queue_rcu(struct rcu_head *rcu_head)
+{
+	struct request_queue *q = container_of(rcu_head, struct request_queue,
+					       rcu_head);
+	kmem_cache_free(blk_requestq_cachep, q);
+}
+
 /**
  * blk_release_queue: - release a &struct request_queue when it is no longer needed
  * @kobj:    the kobj belonging to the request queue to be released
@@ -538,7 +545,7 @@ static void blk_release_queue(struct kobject *kobj)
 	bdi_destroy(&q->backing_dev_info);
 
 	ida_simple_remove(&blk_queue_ida, q->id);
-	kmem_cache_free(blk_requestq_cachep, q);
+	call_rcu(&q->rcu_head, blk_free_queue_rcu);
 }
 
 static const struct sysfs_ops queue_sysfs_ops = {
diff --git a/block/cfq-iosched.c b/block/cfq-iosched.c
index e62e9205b80..b66365b6ba7 100644
--- a/block/cfq-iosched.c
+++ b/block/cfq-iosched.c
@@ -85,7 +85,6 @@ struct cfq_rb_root {
 	struct rb_root rb;
 	struct rb_node *left;
 	unsigned count;
-	unsigned total_weight;
 	u64 min_vdisktime;
 	struct cfq_ttime ttime;
 };
@@ -155,7 +154,7 @@ struct cfq_queue {
  * First index in the service_trees.
  * IDLE is handled separately, so it has negative index
  */
-enum wl_prio_t {
+enum wl_class_t {
 	BE_WORKLOAD = 0,
 	RT_WORKLOAD = 1,
 	IDLE_WORKLOAD = 2,
@@ -223,10 +222,45 @@ struct cfq_group {
 
 	/* group service_tree key */
 	u64 vdisktime;
+
+	/*
+	 * The number of active cfqgs and sum of their weights under this
+	 * cfqg.  This covers this cfqg's leaf_weight and all children's
+	 * weights, but does not cover weights of further descendants.
+	 *
+	 * If a cfqg is on the service tree, it's active.  An active cfqg
+	 * also activates its parent and contributes to the children_weight
+	 * of the parent.
+	 */
+	int nr_active;
+	unsigned int children_weight;
+
+	/*
+	 * vfraction is the fraction of vdisktime that the tasks in this
+	 * cfqg are entitled to.  This is determined by compounding the
+	 * ratios walking up from this cfqg to the root.
+	 *
+	 * It is in fixed point w/ CFQ_SERVICE_SHIFT and the sum of all
+	 * vfractions on a service tree is approximately 1.  The sum may
+	 * deviate a bit due to rounding errors and fluctuations caused by
+	 * cfqgs entering and leaving the service tree.
+	 */
+	unsigned int vfraction;
+
+	/*
+	 * There are two weights - (internal) weight is the weight of this
+	 * cfqg against the sibling cfqgs.  leaf_weight is the wight of
+	 * this cfqg against the child cfqgs.  For the root cfqg, both
+	 * weights are kept in sync for backward compatibility.
+	 */
 	unsigned int weight;
 	unsigned int new_weight;
 	unsigned int dev_weight;
 
+	unsigned int leaf_weight;
+	unsigned int new_leaf_weight;
+	unsigned int dev_leaf_weight;
+
 	/* number of cfqq currently on this group */
 	int nr_cfqq;
 
@@ -248,14 +282,15 @@ struct cfq_group {
 	struct cfq_rb_root service_trees[2][3];
 	struct cfq_rb_root service_tree_idle;
 
-	unsigned long saved_workload_slice;
-	enum wl_type_t saved_workload;
-	enum wl_prio_t saved_serving_prio;
+	unsigned long saved_wl_slice;
+	enum wl_type_t saved_wl_type;
+	enum wl_class_t saved_wl_class;
 
 	/* number of requests that are on the dispatch list or inside driver */
 	int dispatched;
 	struct cfq_ttime ttime;
-	struct cfqg_stats stats;
+	struct cfqg_stats stats;	/* stats for this cfqg */
+	struct cfqg_stats dead_stats;	/* stats pushed from dead children */
 };
 
 struct cfq_io_cq {
@@ -280,8 +315,8 @@ struct cfq_data {
 	/*
 	 * The priority currently being served
 	 */
-	enum wl_prio_t serving_prio;
-	enum wl_type_t serving_type;
+	enum wl_class_t serving_wl_class;
+	enum wl_type_t serving_wl_type;
 	unsigned long workload_expires;
 	struct cfq_group *serving_group;
 
@@ -353,17 +388,17 @@ struct cfq_data {
 
 static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd);
 
-static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg,
-					    enum wl_prio_t prio,
+static struct cfq_rb_root *st_for(struct cfq_group *cfqg,
+					    enum wl_class_t class,
 					    enum wl_type_t type)
 {
 	if (!cfqg)
 		return NULL;
 
-	if (prio == IDLE_WORKLOAD)
+	if (class == IDLE_WORKLOAD)
 		return &cfqg->service_tree_idle;
 
-	return &cfqg->service_trees[prio][type];
+	return &cfqg->service_trees[class][type];
 }
 
 enum cfqq_state_flags {
@@ -502,7 +537,7 @@ static void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg)
 {
 	struct cfqg_stats *stats = &cfqg->stats;
 
-	if (blkg_rwstat_sum(&stats->queued))
+	if (blkg_rwstat_total(&stats->queued))
 		return;
 
 	/*
@@ -546,7 +581,7 @@ static void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg)
 	struct cfqg_stats *stats = &cfqg->stats;
 
 	blkg_stat_add(&stats->avg_queue_size_sum,
-		      blkg_rwstat_sum(&stats->queued));
+		      blkg_rwstat_total(&stats->queued));
 	blkg_stat_add(&stats->avg_queue_size_samples, 1);
 	cfqg_stats_update_group_wait_time(stats);
 }
@@ -572,6 +607,13 @@ static inline struct cfq_group *blkg_to_cfqg(struct blkcg_gq *blkg)
 	return pd_to_cfqg(blkg_to_pd(blkg, &blkcg_policy_cfq));
 }
 
+static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg)
+{
+	struct blkcg_gq *pblkg = cfqg_to_blkg(cfqg)->parent;
+
+	return pblkg ? blkg_to_cfqg(pblkg) : NULL;
+}
+
 static inline void cfqg_get(struct cfq_group *cfqg)
 {
 	return blkg_get(cfqg_to_blkg(cfqg));
@@ -586,8 +628,9 @@ static inline void cfqg_put(struct cfq_group *cfqg)
 	char __pbuf[128];						\
 									\
 	blkg_path(cfqg_to_blkg((cfqq)->cfqg), __pbuf, sizeof(__pbuf));	\
-	blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \
-			  cfq_cfqq_sync((cfqq)) ? 'S' : 'A',		\
+	blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c %s " fmt, (cfqq)->pid, \
+			cfq_cfqq_sync((cfqq)) ? 'S' : 'A',		\
+			cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\
 			  __pbuf, ##args);				\
 } while (0)
 
@@ -646,11 +689,9 @@ static inline void cfqg_stats_update_completion(struct cfq_group *cfqg,
 				io_start_time - start_time);
 }
 
-static void cfq_pd_reset_stats(struct blkcg_gq *blkg)
+/* @stats = 0 */
+static void cfqg_stats_reset(struct cfqg_stats *stats)
 {
-	struct cfq_group *cfqg = blkg_to_cfqg(blkg);
-	struct cfqg_stats *stats = &cfqg->stats;
-
 	/* queued stats shouldn't be cleared */
 	blkg_rwstat_reset(&stats->service_bytes);
 	blkg_rwstat_reset(&stats->serviced);
@@ -669,13 +710,58 @@ static void cfq_pd_reset_stats(struct blkcg_gq *blkg)
 #endif
 }
 
+/* @to += @from */
+static void cfqg_stats_merge(struct cfqg_stats *to, struct cfqg_stats *from)
+{
+	/* queued stats shouldn't be cleared */
+	blkg_rwstat_merge(&to->service_bytes, &from->service_bytes);
+	blkg_rwstat_merge(&to->serviced, &from->serviced);
+	blkg_rwstat_merge(&to->merged, &from->merged);
+	blkg_rwstat_merge(&to->service_time, &from->service_time);
+	blkg_rwstat_merge(&to->wait_time, &from->wait_time);
+	blkg_stat_merge(&from->time, &from->time);
+#ifdef CONFIG_DEBUG_BLK_CGROUP
+	blkg_stat_merge(&to->unaccounted_time, &from->unaccounted_time);
+	blkg_stat_merge(&to->avg_queue_size_sum, &from->avg_queue_size_sum);
+	blkg_stat_merge(&to->avg_queue_size_samples, &from->avg_queue_size_samples);
+	blkg_stat_merge(&to->dequeue, &from->dequeue);
+	blkg_stat_merge(&to->group_wait_time, &from->group_wait_time);
+	blkg_stat_merge(&to->idle_time, &from->idle_time);
+	blkg_stat_merge(&to->empty_time, &from->empty_time);
+#endif
+}
+
+/*
+ * Transfer @cfqg's stats to its parent's dead_stats so that the ancestors'
+ * recursive stats can still account for the amount used by this cfqg after
+ * it's gone.
+ */
+static void cfqg_stats_xfer_dead(struct cfq_group *cfqg)
+{
+	stru