1 // SPDX-License-Identifier: GPL-2.0
2 
3 /*
4  * Based on Christian Brauner's clone3() example.
5  * These tests are assuming to be running in the host's
6  * PID namespace.
7  */
8 
9 /* capabilities related code based on selftests/bpf/test_verifier.c */
10 
11 #define _GNU_SOURCE
12 #include <errno.h>
13 #include <linux/types.h>
14 #include <linux/sched.h>
15 #include <stdio.h>
16 #include <stdlib.h>
17 #include <stdbool.h>
18 #include <sys/capability.h>
19 #include <sys/prctl.h>
20 #include <sys/syscall.h>
21 #include <sys/types.h>
22 #include <sys/un.h>
23 #include <sys/wait.h>
24 #include <unistd.h>
25 #include <sched.h>
26 
27 #include "../kselftest_harness.h"
28 #include "clone3_selftests.h"
29 
30 #ifndef MAX_PID_NS_LEVEL
31 #define MAX_PID_NS_LEVEL 32
32 #endif
33 
child_exit(int ret)34 static void child_exit(int ret)
35 {
36 	fflush(stdout);
37 	fflush(stderr);
38 	_exit(ret);
39 }
40 
call_clone3_set_tid(struct __test_metadata * _metadata,pid_t * set_tid,size_t set_tid_size)41 static int call_clone3_set_tid(struct __test_metadata *_metadata,
42 			       pid_t *set_tid, size_t set_tid_size)
43 {
44 	int status;
45 	pid_t pid = -1;
46 
47 	struct __clone_args args = {
48 		.exit_signal = SIGCHLD,
49 		.set_tid = ptr_to_u64(set_tid),
50 		.set_tid_size = set_tid_size,
51 	};
52 
53 	pid = sys_clone3(&args, sizeof(args));
54 	if (pid < 0) {
55 		TH_LOG("%s - Failed to create new process", strerror(errno));
56 		return -errno;
57 	}
58 
59 	if (pid == 0) {
60 		int ret;
61 		char tmp = 0;
62 
63 		TH_LOG("I am the child, my PID is %d (expected %d)", getpid(), set_tid[0]);
64 
65 		if (set_tid[0] != getpid())
66 			child_exit(EXIT_FAILURE);
67 		child_exit(EXIT_SUCCESS);
68 	}
69 
70 	TH_LOG("I am the parent (%d). My child's pid is %d", getpid(), pid);
71 
72 	if (waitpid(pid, &status, 0) < 0) {
73 		TH_LOG("Child returned %s", strerror(errno));
74 		return -errno;
75 	}
76 
77 	if (!WIFEXITED(status))
78 		return -1;
79 
80 	return WEXITSTATUS(status);
81 }
82 
test_clone3_set_tid(struct __test_metadata * _metadata,pid_t * set_tid,size_t set_tid_size)83 static int test_clone3_set_tid(struct __test_metadata *_metadata,
84 			       pid_t *set_tid, size_t set_tid_size)
85 {
86 	int ret;
87 
88 	TH_LOG("[%d] Trying clone3() with CLONE_SET_TID to %d", getpid(), set_tid[0]);
89 	ret = call_clone3_set_tid(_metadata, set_tid, set_tid_size);
90 	TH_LOG("[%d] clone3() with CLONE_SET_TID %d says:%d", getpid(), set_tid[0], ret);
91 	return ret;
92 }
93 
94 struct libcap {
95 	struct __user_cap_header_struct hdr;
96 	struct __user_cap_data_struct data[2];
97 };
98 
set_capability(void)99 static int set_capability(void)
100 {
101 	cap_value_t cap_values[] = { CAP_SETUID, CAP_SETGID };
102 	struct libcap *cap;
103 	int ret = -1;
104 	cap_t caps;
105 
106 	caps = cap_get_proc();
107 	if (!caps) {
108 		perror("cap_get_proc");
109 		return -1;
110 	}
111 
112 	/* Drop all capabilities */
113 	if (cap_clear(caps)) {
114 		perror("cap_clear");
115 		goto out;
116 	}
117 
118 	cap_set_flag(caps, CAP_EFFECTIVE, 2, cap_values, CAP_SET);
119 	cap_set_flag(caps, CAP_PERMITTED, 2, cap_values, CAP_SET);
120 
121 	cap = (struct libcap *) caps;
122 
123 	/* 40 -> CAP_CHECKPOINT_RESTORE */
124 	cap->data[1].effective |= 1 << (40 - 32);
125 	cap->data[1].permitted |= 1 << (40 - 32);
126 
127 	if (cap_set_proc(caps)) {
128 		perror("cap_set_proc");
129 		goto out;
130 	}
131 	ret = 0;
132 out:
133 	if (cap_free(caps))
134 		perror("cap_free");
135 	return ret;
136 }
137 
TEST(clone3_cap_checkpoint_restore)138 TEST(clone3_cap_checkpoint_restore)
139 {
140 	pid_t pid;
141 	int status;
142 	int ret = 0;
143 	pid_t set_tid[1];
144 
145 	test_clone3_supported();
146 
147 	EXPECT_EQ(getuid(), 0)
148 		SKIP(return, "Skipping all tests as non-root");
149 
150 	memset(&set_tid, 0, sizeof(set_tid));
151 
152 	/* Find the current active PID */
153 	pid = fork();
154 	if (pid == 0) {
155 		TH_LOG("Child has PID %d", getpid());
156 		child_exit(EXIT_SUCCESS);
157 	}
158 	ASSERT_GT(waitpid(pid, &status, 0), 0)
159 		TH_LOG("Waiting for child %d failed", pid);
160 
161 	/* After the child has finished, its PID should be free. */
162 	set_tid[0] = pid;
163 
164 	ASSERT_EQ(set_capability(), 0)
165 		TH_LOG("Could not set CAP_CHECKPOINT_RESTORE");
166 
167 	ASSERT_EQ(prctl(PR_SET_KEEPCAPS, 1, 0, 0, 0), 0);
168 
169 	EXPECT_EQ(setgid(65534), 0)
170 		TH_LOG("Failed to setgid(65534)");
171 	ASSERT_EQ(setuid(65534), 0);
172 
173 	set_tid[0] = pid;
174 	/* This would fail without CAP_CHECKPOINT_RESTORE */
175 	ASSERT_EQ(test_clone3_set_tid(_metadata, set_tid, 1), -EPERM);
176 	ASSERT_EQ(set_capability(), 0)
177 		TH_LOG("Could not set CAP_CHECKPOINT_RESTORE");
178 	/* This should work as we have CAP_CHECKPOINT_RESTORE as non-root */
179 	ASSERT_EQ(test_clone3_set_tid(_metadata, set_tid, 1), 0);
180 }
181 
182 TEST_HARNESS_MAIN
183