* [PATCH v3 1/3] cpuset: Allow setscheduler regardless of manipulated task
2023-07-03 17:27 [PATCH v3 0/3] cpuset: Allow setscheduler regardless of manipulated task Michal Koutný
@ 2023-07-03 17:27 ` Michal Koutný
2023-07-03 18:00 ` Waiman Long
2023-07-03 17:27 ` [PATCH v3 2/3] selftests: cgroup: Minor code reorganizations Michal Koutný
` (2 subsequent siblings)
3 siblings, 1 reply; 7+ messages in thread
From: Michal Koutný @ 2023-07-03 17:27 UTC (permalink / raw)
To: linux-kernel, cgroups, linux-kselftest
Cc: Waiman Long, Zefan Li, Tejun Heo, Johannes Weiner, Shuah Khan
When we migrate a task between two cgroups, one of the checks is a
verification whether we can modify task's scheduler settings
(cap_task_setscheduler()).
An implicit migration occurs also when enabling a controller on the
unified hierarchy (think of parent to child migration). The
aforementioned check may be problematic if the caller of the migration
(enabling a controller) has no permissions over migrated tasks.
For instance, a user's cgroup that ends up running a process of a
different user. Although cgroup permissions are configured favorably,
the enablement fails due to the foreign process [1].
Change the behavior by relaxing the permissions check on the unified
hierarchy when no effective change would happen.
This is in accordance with unified hierarchy attachment behavior when
permissions of the source to target cgroups are decisive whereas the
migrated task is opaque (as opposed to more restrictive check in
__cgroup1_procs_write()).
Notice that foreign task's affinity may still be modified if the user
can modify destination cgroup's cpuset attributes
(update_tasks_cpumask() does no permissions check). The permissions
check could thus be skipped on v2 even when affinity changes. Stay
conservative in this patch though.
[1] https://github.com/systemd/systemd/issues/18293#issuecomment-831205649
Signed-off-by: Michal Koutný <mkoutny@suse.com>
---
kernel/cgroup/cpuset.c | 19 ++++++++++++++++---
1 file changed, 16 insertions(+), 3 deletions(-)
diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index 58e6f18f01c1..0a9b860844ca 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -2487,6 +2487,7 @@ static int cpuset_can_attach(struct cgroup_taskset *tset)
struct cgroup_subsys_state *css;
struct cpuset *cs, *oldcs;
struct task_struct *task;
+ bool cpus_updated, mems_updated;
int ret;
/* used later by cpuset_attach() */
@@ -2501,13 +2502,25 @@ static int cpuset_can_attach(struct cgroup_taskset *tset)
if (ret)
goto out_unlock;
+ cpus_updated = !cpumask_equal(cs->effective_cpus, oldcs->effective_cpus);
+ mems_updated = !nodes_equal(cs->effective_mems, oldcs->effective_mems);
+
cgroup_taskset_for_each(task, css, tset) {
ret = task_can_attach(task);
if (ret)
goto out_unlock;
- ret = security_task_setscheduler(task);
- if (ret)
- goto out_unlock;
+
+ /*
+ * Skip rights over task check in v2 when nothing changes,
+ * migration permission derives from hierarchy ownership in
+ * cgroup_procs_write_permission()).
+ */
+ if (!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) ||
+ (cpus_updated || mems_updated)) {
+ ret = security_task_setscheduler(task);
+ if (ret)
+ goto out_unlock;
+ }
if (dl_task(task)) {
cs->nr_migrate_dl_tasks++;
--
2.41.0
^ permalink raw reply related [flat|nested] 7+ messages in thread* [PATCH v3 2/3] selftests: cgroup: Minor code reorganizations
2023-07-03 17:27 [PATCH v3 0/3] cpuset: Allow setscheduler regardless of manipulated task Michal Koutný
2023-07-03 17:27 ` [PATCH v3 1/3] " Michal Koutný
@ 2023-07-03 17:27 ` Michal Koutný
2023-07-03 18:01 ` Waiman Long
2023-07-03 17:27 ` [PATCH v3 3/3] selftests: cgroup: Add cpuset migrations testcase Michal Koutný
2023-07-10 20:30 ` [PATCH v3 0/3] cpuset: Allow setscheduler regardless of manipulated task Tejun Heo
3 siblings, 1 reply; 7+ messages in thread
From: Michal Koutný @ 2023-07-03 17:27 UTC (permalink / raw)
To: linux-kernel, cgroups, linux-kselftest
Cc: Waiman Long, Zefan Li, Tejun Heo, Johannes Weiner, Shuah Khan
No functional change intended, these small changes are merged into one
commit and they serve as a preparation for an upcoming new testcase.
Signed-off-by: Michal Koutný <mkoutny@suse.com>
---
MAINTAINERS | 1 +
tools/testing/selftests/cgroup/cgroup_util.c | 2 ++
tools/testing/selftests/cgroup/cgroup_util.h | 2 ++
tools/testing/selftests/cgroup/test_core.c | 2 +-
tools/testing/selftests/cgroup/test_cpuset_prs.sh | 2 +-
5 files changed, 7 insertions(+), 2 deletions(-)
diff --git a/MAINTAINERS b/MAINTAINERS
index e0976ae2a523..03bec83944c4 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -5260,6 +5260,7 @@ T: git git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup.git
F: Documentation/admin-guide/cgroup-v1/cpusets.rst
F: include/linux/cpuset.h
F: kernel/cgroup/cpuset.c
+F: tools/testing/selftests/cgroup/test_cpuset_prs.sh
CONTROL GROUP - MEMORY RESOURCE CONTROLLER (MEMCG)
M: Johannes Weiner <hannes@cmpxchg.org>
diff --git a/tools/testing/selftests/cgroup/cgroup_util.c b/tools/testing/selftests/cgroup/cgroup_util.c
index e8bbbdb77e0d..0340d4ca8f51 100644
--- a/tools/testing/selftests/cgroup/cgroup_util.c
+++ b/tools/testing/selftests/cgroup/cgroup_util.c
@@ -286,6 +286,8 @@ int cg_destroy(const char *cgroup)
{
int ret;
+ if (!cgroup)
+ return 0;
retry:
ret = rmdir(cgroup);
if (ret && errno == EBUSY) {
diff --git a/tools/testing/selftests/cgroup/cgroup_util.h b/tools/testing/selftests/cgroup/cgroup_util.h
index c92df4e5d395..1df7f202214a 100644
--- a/tools/testing/selftests/cgroup/cgroup_util.h
+++ b/tools/testing/selftests/cgroup/cgroup_util.h
@@ -11,6 +11,8 @@
#define USEC_PER_SEC 1000000L
#define NSEC_PER_SEC 1000000000L
+#define TEST_UID 65534 /* usually nobody, any !root is fine */
+
/*
* Checks if two given values differ by less than err% of their sum.
*/
diff --git a/tools/testing/selftests/cgroup/test_core.c b/tools/testing/selftests/cgroup/test_core.c
index 600123503063..80aa6b2373b9 100644
--- a/tools/testing/selftests/cgroup/test_core.c
+++ b/tools/testing/selftests/cgroup/test_core.c
@@ -683,7 +683,7 @@ static int test_cgcore_thread_migration(const char *root)
*/
static int test_cgcore_lesser_euid_open(const char *root)
{
- const uid_t test_euid = 65534; /* usually nobody, any !root is fine */
+ const uid_t test_euid = TEST_UID;
int ret = KSFT_FAIL;
char *cg_test_a = NULL, *cg_test_b = NULL;
char *cg_test_a_procs = NULL, *cg_test_b_procs = NULL;
diff --git a/tools/testing/selftests/cgroup/test_cpuset_prs.sh b/tools/testing/selftests/cgroup/test_cpuset_prs.sh
index 2b5215cc599f..4afb132e4e4f 100755
--- a/tools/testing/selftests/cgroup/test_cpuset_prs.sh
+++ b/tools/testing/selftests/cgroup/test_cpuset_prs.sh
@@ -10,7 +10,7 @@
skip_test() {
echo "$1"
echo "Test SKIPPED"
- exit 0
+ exit 4 # ksft_skip
}
[[ $(id -u) -eq 0 ]] || skip_test "Test must be run as root!"
--
2.41.0
^ permalink raw reply related [flat|nested] 7+ messages in thread* [PATCH v3 3/3] selftests: cgroup: Add cpuset migrations testcase
2023-07-03 17:27 [PATCH v3 0/3] cpuset: Allow setscheduler regardless of manipulated task Michal Koutný
2023-07-03 17:27 ` [PATCH v3 1/3] " Michal Koutný
2023-07-03 17:27 ` [PATCH v3 2/3] selftests: cgroup: Minor code reorganizations Michal Koutný
@ 2023-07-03 17:27 ` Michal Koutný
2023-07-10 20:30 ` [PATCH v3 0/3] cpuset: Allow setscheduler regardless of manipulated task Tejun Heo
3 siblings, 0 replies; 7+ messages in thread
From: Michal Koutný @ 2023-07-03 17:27 UTC (permalink / raw)
To: linux-kernel, cgroups, linux-kselftest
Cc: Waiman Long, Zefan Li, Tejun Heo, Johannes Weiner, Shuah Khan
Add a separate testfile to verify treating permissions when tasks are
migrated on cgroup v2 hierarchy between cpuset cgroups.
In accordance with v2 design, migration should be allowed based on
delegation boundaries (i.e. cgroup.procs permissions) and does not
depend on the migrated object (i.e. unprivileged process can migrate
another process (even privileged) as long as it remains in the original
dedicated scope).
Signed-off-by: Michal Koutný <mkoutny@suse.com>
---
MAINTAINERS | 1 +
tools/testing/selftests/cgroup/.gitignore | 1 +
tools/testing/selftests/cgroup/Makefile | 2 +
tools/testing/selftests/cgroup/test_cpuset.c | 275 +++++++++++++++++++
4 files changed, 279 insertions(+)
create mode 100644 tools/testing/selftests/cgroup/test_cpuset.c
diff --git a/MAINTAINERS b/MAINTAINERS
index 03bec83944c4..5c55de000ee3 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -5260,6 +5260,7 @@ T: git git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup.git
F: Documentation/admin-guide/cgroup-v1/cpusets.rst
F: include/linux/cpuset.h
F: kernel/cgroup/cpuset.c
+F: tools/testing/selftests/cgroup/test_cpuset.c
F: tools/testing/selftests/cgroup/test_cpuset_prs.sh
CONTROL GROUP - MEMORY RESOURCE CONTROLLER (MEMCG)
diff --git a/tools/testing/selftests/cgroup/.gitignore b/tools/testing/selftests/cgroup/.gitignore
index c4a57e69f749..8443a8d46a1c 100644
--- a/tools/testing/selftests/cgroup/.gitignore
+++ b/tools/testing/selftests/cgroup/.gitignore
@@ -5,4 +5,5 @@ test_freezer
test_kmem
test_kill
test_cpu
+test_cpuset
wait_inotify
diff --git a/tools/testing/selftests/cgroup/Makefile b/tools/testing/selftests/cgroup/Makefile
index 3d263747d2ad..dee0f013c7f4 100644
--- a/tools/testing/selftests/cgroup/Makefile
+++ b/tools/testing/selftests/cgroup/Makefile
@@ -12,6 +12,7 @@ TEST_GEN_PROGS += test_core
TEST_GEN_PROGS += test_freezer
TEST_GEN_PROGS += test_kill
TEST_GEN_PROGS += test_cpu
+TEST_GEN_PROGS += test_cpuset
LOCAL_HDRS += $(selfdir)/clone3/clone3_selftests.h $(selfdir)/pidfd/pidfd.h
@@ -23,3 +24,4 @@ $(OUTPUT)/test_core: cgroup_util.c
$(OUTPUT)/test_freezer: cgroup_util.c
$(OUTPUT)/test_kill: cgroup_util.c
$(OUTPUT)/test_cpu: cgroup_util.c
+$(OUTPUT)/test_cpuset: cgroup_util.c
diff --git a/tools/testing/selftests/cgroup/test_cpuset.c b/tools/testing/selftests/cgroup/test_cpuset.c
new file mode 100644
index 000000000000..b061ed1e05b4
--- /dev/null
+++ b/tools/testing/selftests/cgroup/test_cpuset.c
@@ -0,0 +1,275 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include <linux/limits.h>
+#include <signal.h>
+
+#include "../kselftest.h"
+#include "cgroup_util.h"
+
+static int idle_process_fn(const char *cgroup, void *arg)
+{
+ (void)pause();
+ return 0;
+}
+
+static int do_migration_fn(const char *cgroup, void *arg)
+{
+ int object_pid = (int)(size_t)arg;
+
+ if (setuid(TEST_UID))
+ return EXIT_FAILURE;
+
+ // XXX checking /proc/$pid/cgroup would be quicker than wait
+ if (cg_enter(cgroup, object_pid) ||
+ cg_wait_for_proc_count(cgroup, 1))
+ return EXIT_FAILURE;
+
+ return EXIT_SUCCESS;
+}
+
+static int do_controller_fn(const char *cgroup, void *arg)
+{
+ const char *child = cgroup;
+ const char *parent = arg;
+
+ if (setuid(TEST_UID))
+ return EXIT_FAILURE;
+
+ if (!cg_read_strstr(child, "cgroup.controllers", "cpuset"))
+ return EXIT_FAILURE;
+
+ if (cg_write(parent, "cgroup.subtree_control", "+cpuset"))
+ return EXIT_FAILURE;
+
+ if (cg_read_strstr(child, "cgroup.controllers", "cpuset"))
+ return EXIT_FAILURE;
+
+ if (cg_write(parent, "cgroup.subtree_control", "-cpuset"))
+ return EXIT_FAILURE;
+
+ if (!cg_read_strstr(child, "cgroup.controllers", "cpuset"))
+ return EXIT_FAILURE;
+
+ return EXIT_SUCCESS;
+}
+
+/*
+ * Migrate a process between two sibling cgroups.
+ * The success should only depend on the parent cgroup permissions and not the
+ * migrated process itself (cpuset controller is in place because it uses
+ * security_task_setscheduler() in cgroup v1).
+ *
+ * Deliberately don't set cpuset.cpus in children to avoid definining migration
+ * permissions between two different cpusets.
+ */
+static int test_cpuset_perms_object(const char *root, bool allow)
+{
+ char *parent = NULL, *child_src = NULL, *child_dst = NULL;
+ char *parent_procs = NULL, *child_src_procs = NULL, *child_dst_procs = NULL;
+ const uid_t test_euid = TEST_UID;
+ int object_pid = 0;
+ int ret = KSFT_FAIL;
+
+ parent = cg_name(root, "cpuset_test_0");
+ if (!parent)
+ goto cleanup;
+ parent_procs = cg_name(parent, "cgroup.procs");
+ if (!parent_procs)
+ goto cleanup;
+ if (cg_create(parent))
+ goto cleanup;
+
+ child_src = cg_name(parent, "cpuset_test_1");
+ if (!child_src)
+ goto cleanup;
+ child_src_procs = cg_name(child_src, "cgroup.procs");
+ if (!child_src_procs)
+ goto cleanup;
+ if (cg_create(child_src))
+ goto cleanup;
+
+ child_dst = cg_name(parent, "cpuset_test_2");
+ if (!child_dst)
+ goto cleanup;
+ child_dst_procs = cg_name(child_dst, "cgroup.procs");
+ if (!child_dst_procs)
+ goto cleanup;
+ if (cg_create(child_dst))
+ goto cleanup;
+
+ if (cg_write(parent, "cgroup.subtree_control", "+cpuset"))
+ goto cleanup;
+
+ if (cg_read_strstr(child_src, "cgroup.controllers", "cpuset") ||
+ cg_read_strstr(child_dst, "cgroup.controllers", "cpuset"))
+ goto cleanup;
+
+ /* Enable permissions along src->dst tree path */
+ if (chown(child_src_procs, test_euid, -1) ||
+ chown(child_dst_procs, test_euid, -1))
+ goto cleanup;
+
+ if (allow && chown(parent_procs, test_euid, -1))
+ goto cleanup;
+
+ /* Fork a privileged child as a test object */
+ object_pid = cg_run_nowait(child_src, idle_process_fn, NULL);
+ if (object_pid < 0)
+ goto cleanup;
+
+ /* Carry out migration in a child process that can drop all privileges
+ * (including capabilities), the main process must remain privileged for
+ * cleanup.
+ * Child process's cgroup is irrelevant but we place it into child_dst
+ * as hacky way to pass information about migration target to the child.
+ */
+ if (allow ^ (cg_run(child_dst, do_migration_fn, (void *)(size_t)object_pid) == EXIT_SUCCESS))
+ goto cleanup;
+
+ ret = KSFT_PASS;
+
+cleanup:
+ if (object_pid > 0) {
+ (void)kill(object_pid, SIGTERM);
+ (void)clone_reap(object_pid, WEXITED);
+ }
+
+ cg_destroy(child_dst);
+ free(child_dst_procs);
+ free(child_dst);
+
+ cg_destroy(child_src);
+ free(child_src_procs);
+ free(child_src);
+
+ cg_destroy(parent);
+ free(parent_procs);
+ free(parent);
+
+ return ret;
+}
+
+static int test_cpuset_perms_object_allow(const char *root)
+{
+ return test_cpuset_perms_object(root, true);
+}
+
+static int test_cpuset_perms_object_deny(const char *root)
+{
+ return test_cpuset_perms_object(root, false);
+}
+
+/*
+ * Migrate a process between parent and child implicitely
+ * Implicit migration happens when a controller is enabled/disabled.
+ *
+ */
+static int test_cpuset_perms_subtree(const char *root)
+{
+ char *parent = NULL, *child = NULL;
+ char *parent_procs = NULL, *parent_subctl = NULL, *child_procs = NULL;
+ const uid_t test_euid = TEST_UID;
+ int object_pid = 0;
+ int ret = KSFT_FAIL;
+
+ parent = cg_name(root, "cpuset_test_0");
+ if (!parent)
+ goto cleanup;
+ parent_procs = cg_name(parent, "cgroup.procs");
+ if (!parent_procs)
+ goto cleanup;
+ parent_subctl = cg_name(parent, "cgroup.subtree_control");
+ if (!parent_subctl)
+ goto cleanup;
+ if (cg_create(parent))
+ goto cleanup;
+
+ child = cg_name(parent, "cpuset_test_1");
+ if (!child)
+ goto cleanup;
+ child_procs = cg_name(child, "cgroup.procs");
+ if (!child_procs)
+ goto cleanup;
+ if (cg_create(child))
+ goto cleanup;
+
+ /* Enable permissions as in a delegated subtree */
+ if (chown(parent_procs, test_euid, -1) ||
+ chown(parent_subctl, test_euid, -1) ||
+ chown(child_procs, test_euid, -1))
+ goto cleanup;
+
+ /* Put a privileged child in the subtree and modify controller state
+ * from an unprivileged process, the main process remains privileged
+ * for cleanup.
+ * The unprivileged child runs in subtree too to avoid parent and
+ * internal-node constraing violation.
+ */
+ object_pid = cg_run_nowait(child, idle_process_fn, NULL);
+ if (object_pid < 0)
+ goto cleanup;
+
+ if (cg_run(child, do_controller_fn, parent) != EXIT_SUCCESS)
+ goto cleanup;
+
+ ret = KSFT_PASS;
+
+cleanup:
+ if (object_pid > 0) {
+ (void)kill(object_pid, SIGTERM);
+ (void)clone_reap(object_pid, WEXITED);
+ }
+
+ cg_destroy(child);
+ free(child_procs);
+ free(child);
+
+ cg_destroy(parent);
+ free(parent_subctl);
+ free(parent_procs);
+ free(parent);
+
+ return ret;
+}
+
+
+#define T(x) { x, #x }
+struct cpuset_test {
+ int (*fn)(const char *root);
+ const char *name;
+} tests[] = {
+ T(test_cpuset_perms_object_allow),
+ T(test_cpuset_perms_object_deny),
+ T(test_cpuset_perms_subtree),
+};
+#undef T
+
+int main(int argc, char *argv[])
+{
+ char root[PATH_MAX];
+ int i, ret = EXIT_SUCCESS;
+
+ if (cg_find_unified_root(root, sizeof(root)))
+ ksft_exit_skip("cgroup v2 isn't mounted\n");
+
+ if (cg_read_strstr(root, "cgroup.subtree_control", "cpuset"))
+ if (cg_write(root, "cgroup.subtree_control", "+cpuset"))
+ ksft_exit_skip("Failed to set cpuset controller\n");
+
+ for (i = 0; i < ARRAY_SIZE(tests); i++) {
+ switch (tests[i].fn(root)) {
+ case KSFT_PASS:
+ ksft_test_result_pass("%s\n", tests[i].name);
+ break;
+ case KSFT_SKIP:
+ ksft_test_result_skip("%s\n", tests[i].name);
+ break;
+ default:
+ ret = EXIT_FAILURE;
+ ksft_test_result_fail("%s\n", tests[i].name);
+ break;
+ }
+ }
+
+ return ret;
+}
--
2.41.0
^ permalink raw reply related [flat|nested] 7+ messages in thread