|
| 1 | +// SPDX-License-Identifier: GPL-2.0 |
| 2 | +#define _GNU_SOURCE |
| 3 | +#include <fcntl.h> |
| 4 | +#include <sched.h> |
| 5 | +#include <stdio.h> |
| 6 | +#include <sys/types.h> |
| 7 | +#include <unistd.h> |
| 8 | + |
| 9 | +#include "kselftest_harness.h" |
| 10 | +#include "../pidfd/pidfd.h" |
| 11 | + |
| 12 | +/* |
| 13 | + * Test that a process can become PID 1 (init) in a new PID namespace |
| 14 | + * created via unshare() and joined via setns(). |
| 15 | + * |
| 16 | + * Flow: |
| 17 | + * 1. Parent creates a pipe for synchronization. |
| 18 | + * 2. Parent forks a child. |
| 19 | + * 3. Parent calls unshare(CLONE_NEWPID) to create a new PID namespace. |
| 20 | + * 4. Parent signals the child via the pipe. |
| 21 | + * 5. Child opens parent's /proc/<ppid>/ns/pid_for_children and calls |
| 22 | + * setns(fd, CLONE_NEWPID) to join the new namespace. |
| 23 | + * 6. Child forks a grandchild. |
| 24 | + * 7. Grandchild verifies getpid() == 1. |
| 25 | + */ |
| 26 | +TEST(pidns_init_via_setns) |
| 27 | +{ |
| 28 | + pid_t child, parent_pid; |
| 29 | + int pipe_fd[2]; |
| 30 | + char buf; |
| 31 | + |
| 32 | + if (geteuid()) |
| 33 | + ASSERT_EQ(0, unshare(CLONE_NEWUSER)); |
| 34 | + |
| 35 | + parent_pid = getpid(); |
| 36 | + |
| 37 | + ASSERT_EQ(0, pipe(pipe_fd)); |
| 38 | + |
| 39 | + child = fork(); |
| 40 | + ASSERT_GE(child, 0); |
| 41 | + |
| 42 | + if (child == 0) { |
| 43 | + char path[256]; |
| 44 | + int nsfd; |
| 45 | + pid_t grandchild; |
| 46 | + |
| 47 | + close(pipe_fd[1]); |
| 48 | + |
| 49 | + /* Wait for parent to complete unshare */ |
| 50 | + ASSERT_EQ(1, read_nointr(pipe_fd[0], &buf, 1)); |
| 51 | + close(pipe_fd[0]); |
| 52 | + |
| 53 | + snprintf(path, sizeof(path), |
| 54 | + "/proc/%d/ns/pid_for_children", parent_pid); |
| 55 | + nsfd = open(path, O_RDONLY); |
| 56 | + ASSERT_GE(nsfd, 0); |
| 57 | + |
| 58 | + ASSERT_EQ(0, setns(nsfd, CLONE_NEWPID)); |
| 59 | + close(nsfd); |
| 60 | + |
| 61 | + grandchild = fork(); |
| 62 | + ASSERT_GE(grandchild, 0); |
| 63 | + |
| 64 | + if (grandchild == 0) { |
| 65 | + /* Should be init (PID 1) in the new namespace */ |
| 66 | + if (getpid() != 1) |
| 67 | + _exit(1); |
| 68 | + _exit(0); |
| 69 | + } |
| 70 | + |
| 71 | + ASSERT_EQ(0, wait_for_pid(grandchild)); |
| 72 | + _exit(0); |
| 73 | + } |
| 74 | + |
| 75 | + close(pipe_fd[0]); |
| 76 | + |
| 77 | + ASSERT_EQ(0, unshare(CLONE_NEWPID)); |
| 78 | + |
| 79 | + /* Signal child that the new PID namespace is ready */ |
| 80 | + buf = 0; |
| 81 | + ASSERT_EQ(1, write_nointr(pipe_fd[1], &buf, 1)); |
| 82 | + close(pipe_fd[1]); |
| 83 | + |
| 84 | + ASSERT_EQ(0, wait_for_pid(child)); |
| 85 | +} |
| 86 | + |
| 87 | +/* |
| 88 | + * Similar to pidns_init_via_setns, but: |
| 89 | + * 1. Parent enters a new PID namespace right from the start to be able to |
| 90 | + * later freely use pid 1001 in it. |
| 91 | + * 2. After forking child, parent also calls unshare(CLONE_NEWUSER) |
| 92 | + * before unshare(CLONE_NEWPID) so that new old and new pid namespaces have |
| 93 | + * different user namespace owners. |
| 94 | + * 3. Child uses clone3() with set_tid={1, 1001} instead of fork() and |
| 95 | + * grandchild checks that it gets desired pids . |
| 96 | + * |
| 97 | + * Flow: |
| 98 | + * 1. Test process creates a new PID namespace and forks a wrapper |
| 99 | + * (PID 1 in the outer namespace). |
| 100 | + * 2. Wrapper forks a child. |
| 101 | + * 3. Wrapper calls unshare(CLONE_NEWUSER) + unshare(CLONE_NEWPID) |
| 102 | + * to create an inner PID namespace. |
| 103 | + * 4. Wrapper signals the child via pipe. |
| 104 | + * 5. Child opens wrapper's /proc/<pid>/ns/pid_for_children and calls |
| 105 | + * setns(fd, CLONE_NEWPID) to join the inner namespace. |
| 106 | + * 6. Child calls clone3() with set_tid={1, 1001}. |
| 107 | + * 7. Grandchild verifies its NSpid ends with "1001 1". |
| 108 | + */ |
| 109 | + |
| 110 | +pid_t set_tid[] = {1, 1001}; |
| 111 | + |
| 112 | +static int pidns_init_via_setns_set_tid_grandchild(struct __test_metadata *_metadata) |
| 113 | +{ |
| 114 | + char *line = NULL; |
| 115 | + size_t len = 0; |
| 116 | + int found = 0; |
| 117 | + FILE *gf; |
| 118 | + |
| 119 | + gf = fopen("/proc/self/status", "r"); |
| 120 | + ASSERT_NE(gf, NULL); |
| 121 | + |
| 122 | + while (getline(&line, &len, gf) != -1) { |
| 123 | + if (strncmp(line, "NSpid:", 6) != 0) |
| 124 | + continue; |
| 125 | + |
| 126 | + for (int i = 0; i < 2; i++) { |
| 127 | + char *last = strrchr(line, '\t'); |
| 128 | + pid_t pid; |
| 129 | + |
| 130 | + ASSERT_NE(last, NULL); |
| 131 | + ASSERT_EQ(sscanf(last, "%d", &pid), 1); |
| 132 | + ASSERT_EQ(pid, set_tid[i]); |
| 133 | + *last = '\0'; |
| 134 | + } |
| 135 | + |
| 136 | + found = true; |
| 137 | + break; |
| 138 | + } |
| 139 | + |
| 140 | + free(line); |
| 141 | + fclose(gf); |
| 142 | + ASSERT_TRUE(found); |
| 143 | + return 0; |
| 144 | +} |
| 145 | + |
| 146 | +static int pidns_init_via_setns_set_tid_child(struct __test_metadata *_metadata, |
| 147 | + pid_t parent_pid, int pipe_fd[2]) |
| 148 | +{ |
| 149 | + struct __clone_args args = { |
| 150 | + .exit_signal = SIGCHLD, |
| 151 | + .set_tid = ptr_to_u64(set_tid), |
| 152 | + .set_tid_size = 2, |
| 153 | + }; |
| 154 | + pid_t grandchild; |
| 155 | + char path[256]; |
| 156 | + char buf; |
| 157 | + int nsfd; |
| 158 | + |
| 159 | + close(pipe_fd[1]); |
| 160 | + |
| 161 | + ASSERT_EQ(1, read_nointr(pipe_fd[0], &buf, 1)); |
| 162 | + close(pipe_fd[0]); |
| 163 | + |
| 164 | + snprintf(path, sizeof(path), |
| 165 | + "/proc/%d/ns/pid_for_children", parent_pid); |
| 166 | + nsfd = open(path, O_RDONLY); |
| 167 | + ASSERT_GE(nsfd, 0); |
| 168 | + |
| 169 | + ASSERT_EQ(0, setns(nsfd, CLONE_NEWPID)); |
| 170 | + close(nsfd); |
| 171 | + |
| 172 | + grandchild = sys_clone3(&args, sizeof(args)); |
| 173 | + ASSERT_GE(grandchild, 0); |
| 174 | + |
| 175 | + if (grandchild == 0) |
| 176 | + _exit(pidns_init_via_setns_set_tid_grandchild(_metadata)); |
| 177 | + |
| 178 | + ASSERT_EQ(0, wait_for_pid(grandchild)); |
| 179 | + return 0; |
| 180 | +} |
| 181 | + |
| 182 | +static int pidns_init_via_setns_set_tid_wrapper(struct __test_metadata *_metadata) |
| 183 | +{ |
| 184 | + int pipe_fd[2]; |
| 185 | + pid_t child, parent_pid; |
| 186 | + char buf; |
| 187 | + FILE *f; |
| 188 | + |
| 189 | + /* |
| 190 | + * We are PID 1 inside the new namespace, but /proc is |
| 191 | + * mounted from the host. Read our host-visible PID so |
| 192 | + * the child can reach our pid_for_children via /proc. |
| 193 | + */ |
| 194 | + f = fopen("/proc/self/stat", "r"); |
| 195 | + ASSERT_NE(f, NULL); |
| 196 | + ASSERT_EQ(fscanf(f, "%d", &parent_pid), 1); |
| 197 | + ASSERT_EQ(0, pipe(pipe_fd)); |
| 198 | + |
| 199 | + child = fork(); |
| 200 | + ASSERT_GE(child, 0); |
| 201 | + |
| 202 | + if (child == 0) |
| 203 | + _exit(pidns_init_via_setns_set_tid_child(_metadata, parent_pid, pipe_fd)); |
| 204 | + |
| 205 | + close(pipe_fd[0]); |
| 206 | + |
| 207 | + ASSERT_EQ(0, unshare(CLONE_NEWUSER)); |
| 208 | + ASSERT_EQ(0, unshare(CLONE_NEWPID)); |
| 209 | + |
| 210 | + buf = 0; |
| 211 | + ASSERT_EQ(1, write_nointr(pipe_fd[1], &buf, 1)); |
| 212 | + close(pipe_fd[1]); |
| 213 | + |
| 214 | + ASSERT_EQ(0, wait_for_pid(child)); |
| 215 | + |
| 216 | + fclose(f); |
| 217 | + return 0; |
| 218 | +} |
| 219 | + |
| 220 | +TEST(pidns_init_via_setns_set_tid) |
| 221 | +{ |
| 222 | + pid_t wrapper; |
| 223 | + |
| 224 | + if (geteuid()) |
| 225 | + SKIP(return, "This test needs root to run!"); |
| 226 | + |
| 227 | + ASSERT_EQ(0, unshare(CLONE_NEWPID)); |
| 228 | + |
| 229 | + wrapper = fork(); |
| 230 | + ASSERT_GE(wrapper, 0); |
| 231 | + |
| 232 | + if (wrapper == 0) |
| 233 | + _exit(pidns_init_via_setns_set_tid_wrapper(_metadata)); |
| 234 | + |
| 235 | + ASSERT_EQ(0, wait_for_pid(wrapper)); |
| 236 | +} |
| 237 | + |
| 238 | +TEST_HARNESS_MAIN |
0 commit comments