sched_ext: Maintain an online cid mask in the scheduler arenagrafted

Schedulers on the default cid mapping treat [0, nr_online_cids) as the
online set and restart on hotplug. Schedulers that install their own mapping
with scx_bpf_cid_override() have no way to learn which cids are online: the
count no longer identifies members and the CPU-form cpumask is unusable from
cid programs. This is an obvious hole in the cid API.

Add scx_bpf_online_cmask(), a kernel-maintained cmask in the scheduler's
arena, allocated alongside the per-CPU scratch masks and populated after the
cid mapping is finalized and before ops.init(), for child schedulers too.
The pointer stays valid through ops.exit() with no reference to take. It is
the arena offset as a void pointer, the same form struct_ops arena arguments
arrive in. The verifier types the void return as a scalar for the program's
arena cast.

The mask follows the SCX hotplug notifications: seeded from cpu_active_mask
and updated before ops.cid_online/offline() runs, so it lags cpu_online_mask
only inside a hotplug transition. Updates walk the scheduler list under the
lock that also serializes unlinking. Reads are live, not atomic snapshots.
Root initialization excludes hotplug.

v2: Reworded the getter kerneldoc (Andrea Righi).

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>

1 files changed, 88 insertions, 0 deletions

diff --git a/include/net/cls_cgroup.h b/include/net/cls_cgroup.h
new file mode 100644
index 000000000..668aeee9b
--- /dev/null
+++ b/include/net/cls_cgroup.h
@@ -0,0 +1,88 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * cls_cgroup.h			Control Group Classifier
+ *
+ * Authors:	Thomas Graf <tgraf@suug.ch>
+ */
+
+#ifndef _NET_CLS_CGROUP_H
+#define _NET_CLS_CGROUP_H
+
+#include <linux/cgroup.h>
+#include <linux/hardirq.h>
+#include <linux/rcupdate.h>
+#include <net/sock.h>
+#include <net/inet_sock.h>
+
+#ifdef CONFIG_CGROUP_NET_CLASSID
+struct cgroup_cls_state {
+	struct cgroup_subsys_state css;
+	u32 classid;
+};
+
+struct cgroup_cls_state *task_cls_state(struct task_struct *p);
+
+static inline u32 task_cls_classid(struct task_struct *p)
+{
+	u32 classid;
+
+	if (in_interrupt())
+		return 0;
+
+	rcu_read_lock();
+	classid = container_of(task_css(p, net_cls_cgrp_id),
+			       struct cgroup_cls_state, css)->classid;
+	rcu_read_unlock();
+
+	return classid;
+}
+
+static inline void sock_update_classid(struct sock_cgroup_data *skcd)
+{
+	u32 classid;
+
+	classid = task_cls_classid(current);
+	sock_cgroup_set_classid(skcd, classid);
+}
+
+static inline u32 __task_get_classid(struct task_struct *task)
+{
+	return task_cls_state(task)->classid;
+}
+
+static inline u32 task_get_classid(const struct sk_buff *skb)
+{
+	u32 classid = __task_get_classid(current);
+
+	/* Due to the nature of the classifier it is required to ignore all
+	 * packets originating from softirq context as accessing `current'
+	 * would lead to false results.
+	 *
+	 * This test assumes that all callers of dev_queue_xmit() explicitly
+	 * disable bh. Knowing this, it is possible to detect softirq based
+	 * calls by looking at the number of nested bh disable calls because
+	 * softirqs always disables bh.
+	 */
+	if (softirq_count()) {
+		struct sock *sk = skb_to_full_sk(skb);
+
+		/* If there is an sock_cgroup_classid we'll use that. */
+		if (!sk || !sk_fullsock(sk))
+			return 0;
+
+		classid = sock_cgroup_classid(&sk->sk_cgrp_data);
+	}
+
+	return classid;
+}
+#else /* !CONFIG_CGROUP_NET_CLASSID */
+static inline void sock_update_classid(struct sock_cgroup_data *skcd)
+{
+}
+
+static inline u32 task_get_classid(const struct sk_buff *skb)
+{
+	return 0;
+}
+#endif /* CONFIG_CGROUP_NET_CLASSID */
+#endif  /* _NET_CLS_CGROUP_H */