Merge branch 'relax-allowlist-for-open-coded-css_task-iter'

Chuyi Zhou says:

====================
Relax allowlist for open-coded css_task iter

Hi,
The patchset aims to relax the allowlist for open-coded css_task iter
suggested by Alexei[1].

Please see individual patches for more details. And comments are always
welcome.

Patch summary:
 * Patch #1: Relax the allowlist and let css_task iter can be used in
   bpf iters and any sleepable progs.
 * Patch #2: Add a test in cgroup_iters.c which demonstrates how
   css_task iters can be combined with cgroup iter.
 * Patch #3: Add a test to prove css_task iter can be used in normal
 * sleepable progs.
link[1]:https://lore.kernel.org/lkml/CAADnVQKafk_junRyE=-FVAik4hjTRDtThymYGEL8hGTuYoOGpA@mail.gmail.com/
---

Changes in v2:
 * Fix the incorrect logic in check_css_task_iter_allowlist. Use
   expected_attach_type to check whether we are using bpf_iters.
 * Link to v1:https://lore.kernel.org/bpf/20231022154527.229117-1-zhouchuyi@bytedance.com/T/#m946f9cde86b44a13265d9a44c5738a711eb578fd
Changes in v3:
 * Add a testcase to prove css_task can be used in fentry.s
 * Link to v2:https://lore.kernel.org/bpf/20231024024240.42790-1-zhouchuyi@bytedance.com/T/#m14a97041ff56c2df21bc0149449abd275b73f6a3
Changes in v4:
 * Add Yonghong's ack for patch #1 and patch #2.
 * Solve Yonghong's comments for patch #2
 * Move prog 'iter_css_task_for_each_sleep' from iters_task_failure.c to
   iters_css_task.c. Use RUN_TESTS to prove we can load this prog.
 * Link to v3:https://lore.kernel.org/bpf/20231025075914.30979-1-zhouchuyi@bytedance.com/T/#m3200d8ad29af4ffab97588e297361d0a45d7585d

---
====================

Link: https://lore.kernel.org/r/20231031050438.93297-1-zhouchuyi@bytedance.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

kernel/bpf/verifier.c

@@ -11402,17 +11402,25 @@ static int process_kf_arg_ptr_to_rbtree_node(struct bpf_verifier_env *env,
 						  &meta->arg_rbtree_root.field);
 }
 
+/*
+ * css_task iter allowlist is needed to avoid dead locking on css_set_lock.
+ * LSM hooks and iters (both sleepable and non-sleepable) are safe.
+ * Any sleepable progs are also safe since bpf_check_attach_target() enforce
+ * them can only be attached to some specific hook points.
+ */
 static bool check_css_task_iter_allowlist(struct bpf_verifier_env *env)
 {
 	enum bpf_prog_type prog_type = resolve_prog_type(env->prog);
 
 	switch (prog_type) {
 	case BPF_PROG_TYPE_LSM:
 		return true;
-	case BPF_TRACE_ITER:
-		return env->prog->aux->sleepable;
+	case BPF_PROG_TYPE_TRACING:
+		if (env->prog->expected_attach_type == BPF_TRACE_ITER)
+			return true;
+		fallthrough;
 	default:
-		return false;
+		return env->prog->aux->sleepable;
 	}
 }
 
@@ -11671,7 +11679,7 @@ static int check_kfunc_args(struct bpf_verifier_env *env, struct bpf_kfunc_call_
 		case KF_ARG_PTR_TO_ITER:
 			if (meta->func_id == special_kfunc_list[KF_bpf_iter_css_task_new]) {
 				if (!check_css_task_iter_allowlist(env)) {
-					verbose(env, "css_task_iter is only allowed in bpf_lsm and bpf iter-s\n");
+					verbose(env, "css_task_iter is only allowed in bpf_lsm, bpf_iter and sleepable progs\n");
 					return -EINVAL;
 				}
 			}

tools/testing/selftests/bpf/prog_tests/cgroup_iter.c

@@ -4,6 +4,7 @@
 #include <test_progs.h>
 #include <bpf/libbpf.h>
 #include <bpf/btf.h>
+#include "iters_css_task.skel.h"
 #include "cgroup_iter.skel.h"
 #include "cgroup_helpers.h"
 
@@ -263,6 +264,35 @@ static void test_walk_dead_self_only(struct cgroup_iter *skel)
 	close(cgrp_fd);
 }
 
+static void test_walk_self_only_css_task(void)
+{
+	struct iters_css_task *skel;
+	int err;
+
+	skel = iters_css_task__open();
+	if (!ASSERT_OK_PTR(skel, "skel_open"))
+		return;
+
+	bpf_program__set_autoload(skel->progs.cgroup_id_printer, true);
+
+	err = iters_css_task__load(skel);
+	if (!ASSERT_OK(err, "skel_load"))
+		goto cleanup;
+
+	err = join_cgroup(cg_path[CHILD2]);
+	if (!ASSERT_OK(err, "join_cgroup"))
+		goto cleanup;
+
+	skel->bss->target_pid = getpid();
+	snprintf(expected_output, sizeof(expected_output),
+		PROLOGUE "%8llu\n" EPILOGUE, cg_id[CHILD2]);
+	read_from_cgroup_iter(skel->progs.cgroup_id_printer, cg_fd[CHILD2],
+		BPF_CGROUP_ITER_SELF_ONLY, "test_walk_self_only_css_task");
+	ASSERT_EQ(skel->bss->css_task_cnt, 1, "css_task_cnt");
+cleanup:
+	iters_css_task__destroy(skel);
+}
+
 void test_cgroup_iter(void)
 {
 	struct cgroup_iter *skel = NULL;
@@ -293,6 +323,9 @@ void test_cgroup_iter(void)
 		test_walk_self_only(skel);
 	if (test__start_subtest("cgroup_iter__dead_self_only"))
 		test_walk_dead_self_only(skel);
+	if (test__start_subtest("cgroup_iter__self_only_css_task"))
+		test_walk_self_only_css_task();
+
 out:
 	cgroup_iter__destroy(skel);
 	cleanup_cgroups();

tools/testing/selftests/bpf/prog_tests/iters.c

@@ -294,6 +294,7 @@ void test_iters(void)
 	RUN_TESTS(iters_state_safety);
 	RUN_TESTS(iters_looping);
 	RUN_TESTS(iters);
+	RUN_TESTS(iters_css_task);
 
 	if (env.has_testmod)
 		RUN_TESTS(iters_testmod_seq);

tools/testing/selftests/bpf/progs/iters_css_task.c

@@ -10,6 +10,7 @@
 
 char _license[] SEC("license") = "GPL";
 
+struct cgroup *bpf_cgroup_acquire(struct cgroup *p) __ksym;
 struct cgroup *bpf_cgroup_from_id(u64 cgid) __ksym;
 void bpf_cgroup_release(struct cgroup *p) __ksym;
 
@@ -45,3 +46,65 @@ int BPF_PROG(iter_css_task_for_each, struct vm_area_struct *vma,
 
 	return -EPERM;
 }
+
+static inline u64 cgroup_id(struct cgroup *cgrp)
+{
+	return cgrp->kn->id;
+}
+
+SEC("?iter/cgroup")
+int cgroup_id_printer(struct bpf_iter__cgroup *ctx)
+{
+	struct seq_file *seq = ctx->meta->seq;
+	struct cgroup *cgrp, *acquired;
+	struct cgroup_subsys_state *css;
+	struct task_struct *task;
+	u64 cgrp_id;
+
+	cgrp = ctx->cgroup;
+
+	/* epilogue */
+	if (cgrp == NULL) {
+		BPF_SEQ_PRINTF(seq, "epilogue\n");
+		return 0;
+	}
+
+	/* prologue */
+	if (ctx->meta->seq_num == 0)
+		BPF_SEQ_PRINTF(seq, "prologue\n");
+
+	cgrp_id = cgroup_id(cgrp);
+
+	BPF_SEQ_PRINTF(seq, "%8llu\n", cgrp_id);
+
+	acquired = bpf_cgroup_from_id(cgrp_id);
+	if (!acquired)
+		return 0;
+	css = &acquired->self;
+	css_task_cnt = 0;
+	bpf_for_each(css_task, task, css, CSS_TASK_ITER_PROCS) {
+		if (task->pid == target_pid)
+			css_task_cnt++;
+	}
+	bpf_cgroup_release(acquired);
+	return 0;
+}
+
+SEC("?fentry.s/" SYS_PREFIX "sys_getpgid")
+int BPF_PROG(iter_css_task_for_each_sleep)
+{
+	u64 cgrp_id = bpf_get_current_cgroup_id();
+	struct cgroup *cgrp = bpf_cgroup_from_id(cgrp_id);
+	struct cgroup_subsys_state *css;
+	struct task_struct *task;
+
+	if (cgrp == NULL)
+		return 0;
+	css = &cgrp->self;
+
+	bpf_for_each(css_task, task, css, CSS_TASK_ITER_PROCS) {
+
+	}
+	bpf_cgroup_release(cgrp);
+	return 0;
+}

tools/testing/selftests/bpf/progs/iters_task_failure.c

@@ -84,8 +84,8 @@ int BPF_PROG(iter_css_lock_and_unlock)
 	return 0;
 }
 
-SEC("?fentry.s/" SYS_PREFIX "sys_getpgid")
-__failure __msg("css_task_iter is only allowed in bpf_lsm and bpf iter-s")
+SEC("?fentry/" SYS_PREFIX "sys_getpgid")
+__failure __msg("css_task_iter is only allowed in bpf_lsm, bpf_iter and sleepable progs")
 int BPF_PROG(iter_css_task_for_each)
 {
 	u64 cg_id = bpf_get_current_cgroup_id();