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>

4 files changed, 170 insertions, 0 deletions

diff --git a/tools/testing/selftests/rlimits/.gitignore b/tools/testing/selftests/rlimits/.gitignore
new file mode 100644
index 000000000..091021f25
--- /dev/null
+++ b/tools/testing/selftests/rlimits/.gitignore
@@ -0,0 +1,2 @@
+# SPDX-License-Identifier: GPL-2.0-only
+rlimits-per-userns
diff --git a/tools/testing/selftests/rlimits/Makefile b/tools/testing/selftests/rlimits/Makefile
new file mode 100644
index 000000000..03aadb406
--- /dev/null
+++ b/tools/testing/selftests/rlimits/Makefile
@@ -0,0 +1,6 @@
+# SPDX-License-Identifier: GPL-2.0-or-later
+
+CFLAGS += -Wall -O2 -g
+TEST_GEN_PROGS := rlimits-per-userns
+
+include ../lib.mk
diff --git a/tools/testing/selftests/rlimits/config b/tools/testing/selftests/rlimits/config
new file mode 100644
index 000000000..416bd53ce
--- /dev/null
+++ b/tools/testing/selftests/rlimits/config
@@ -0,0 +1 @@
+CONFIG_USER_NS=y
diff --git a/tools/testing/selftests/rlimits/rlimits-per-userns.c b/tools/testing/selftests/rlimits/rlimits-per-userns.c
new file mode 100644
index 000000000..26dc949e9
--- /dev/null
+++ b/tools/testing/selftests/rlimits/rlimits-per-userns.c
@@ -0,0 +1,161 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Author: Alexey Gladkov <gladkov.alexey@gmail.com>
+ */
+#define _GNU_SOURCE
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <sys/time.h>
+#include <sys/resource.h>
+#include <sys/prctl.h>
+#include <sys/stat.h>
+
+#include <unistd.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include <sched.h>
+#include <signal.h>
+#include <limits.h>
+#include <fcntl.h>
+#include <errno.h>
+#include <err.h>
+
+#define NR_CHILDS 2
+
+static char *service_prog;
+static uid_t user   = 60000;
+static uid_t group  = 60000;
+
+static void setrlimit_nproc(rlim_t n)
+{
+	pid_t pid = getpid();
+	struct rlimit limit = {
+		.rlim_cur = n,
+		.rlim_max = n
+	};
+
+	warnx("(pid=%d): Setting RLIMIT_NPROC=%ld", pid, n);
+
+	if (setrlimit(RLIMIT_NPROC, &limit) < 0)
+		err(EXIT_FAILURE, "(pid=%d): setrlimit(RLIMIT_NPROC)", pid);
+}
+
+static pid_t fork_child(void)
+{
+	pid_t pid = fork();
+
+	if (pid < 0)
+		err(EXIT_FAILURE, "fork");
+
+	if (pid > 0)
+		return pid;
+
+	pid = getpid();
+
+	warnx("(pid=%d): New process starting ...", pid);
+
+	if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
+		err(EXIT_FAILURE, "(pid=%d): prctl(PR_SET_PDEATHSIG)", pid);
+
+	signal(SIGUSR1, SIG_DFL);
+
+	warnx("(pid=%d): Changing to uid=%d, gid=%d", pid, user, group);
+
+	if (setgid(group) < 0)
+		err(EXIT_FAILURE, "(pid=%d): setgid(%d)", pid, group);
+	if (setuid(user) < 0)
+		err(EXIT_FAILURE, "(pid=%d): setuid(%d)", pid, user);
+
+	warnx("(pid=%d): Service running ...", pid);
+
+	warnx("(pid=%d): Unshare user namespace", pid);
+	if (unshare(CLONE_NEWUSER) < 0)
+		err(EXIT_FAILURE, "unshare(CLONE_NEWUSER)");
+
+	char *const argv[] = { "service", NULL };
+	char *const envp[] = { "I_AM_SERVICE=1", NULL };
+
+	warnx("(pid=%d): Executing real service ...", pid);
+
+	execve(service_prog, argv, envp);
+	err(EXIT_FAILURE, "(pid=%d): execve", pid);
+}
+
+int main(int argc, char **argv)
+{
+	size_t i;
+	pid_t child[NR_CHILDS];
+	int wstatus[NR_CHILDS];
+	int childs = NR_CHILDS;
+	pid_t pid;
+
+	if (getenv("I_AM_SERVICE")) {
+		pause();
+		exit(EXIT_SUCCESS);
+	}
+
+	service_prog = argv[0];
+	pid = getpid();
+
+	warnx("(pid=%d) Starting testcase", pid);
+
+	/*
+	 * This rlimit is not a problem for root because it can be exceeded.
+	 */
+	setrlimit_nproc(1);
+
+	for (i = 0; i < NR_CHILDS; i++) {
+		child[i] = fork_child();
+		wstatus[i] = 0;
+		usleep(250000);
+	}
+
+	while (1) {
+		for (i = 0; i < NR_CHILDS; i++) {
+			if (child[i] <= 0)
+				continue;
+
+			errno = 0;
+			pid_t ret = waitpid(child[i], &wstatus[i], WNOHANG);
+
+			if (!ret || (!WIFEXITED(wstatus[i]) && !WIFSIGNALED(wstatus[i])))
+				continue;
+
+			if (ret < 0 && errno != ECHILD)
+				warn("(pid=%d): waitpid(%d)", pid, child[i]);
+
+			child[i] *= -1;
+			childs -= 1;
+		}
+
+		if (!childs)
+			break;
+
+		usleep(250000);
+
+		for (i = 0; i < NR_CHILDS; i++) {
+			if (child[i] <= 0)
+				continue;
+			kill(child[i], SIGUSR1);
+		}
+	}
+
+	for (i = 0; i < NR_CHILDS; i++) {
+		if (WIFEXITED(wstatus[i]))
+			warnx("(pid=%d): pid %d exited, status=%d",
+				pid, -child[i], WEXITSTATUS(wstatus[i]));
+		else if (WIFSIGNALED(wstatus[i]))
+			warnx("(pid=%d): pid %d killed by signal %d",
+				pid, -child[i], WTERMSIG(wstatus[i]));
+
+		if (WIFSIGNALED(wstatus[i]) && WTERMSIG(wstatus[i]) == SIGUSR1)
+			continue;
+
+		warnx("(pid=%d): Test failed", pid);
+		exit(EXIT_FAILURE);
+	}
+
+	warnx("(pid=%d): Test passed", pid);
+	exit(EXIT_SUCCESS);
+}