9 #include <sys/sysmacros.h>
10 #include <sys/types.h>
12 #include <sys/socket.h>
15 #include <sys/syscall.h>
17 #include <sys/ioctl.h>
18 #include <sys/ptrace.h>
19 #include <sys/mount.h>
20 #include <linux/limits.h>
21 #include <linux/filter.h>
22 #include <linux/seccomp.h>
24 #define ARRAY_SIZE(x) (sizeof(x) / sizeof(*(x)))
26 static int seccomp(unsigned int op, unsigned int flags, void *args)
29 return syscall(__NR_seccomp, op, flags, args);
32 static int send_fd(int sock, int fd)
34 struct msghdr msg = {};
36 char buf[CMSG_SPACE(sizeof(int))] = {0}, c = 'c';
44 msg.msg_control = buf;
45 msg.msg_controllen = sizeof(buf);
46 cmsg = CMSG_FIRSTHDR(&msg);
47 cmsg->cmsg_level = SOL_SOCKET;
48 cmsg->cmsg_type = SCM_RIGHTS;
49 cmsg->cmsg_len = CMSG_LEN(sizeof(int));
50 *((int *)CMSG_DATA(cmsg)) = fd;
51 msg.msg_controllen = cmsg->cmsg_len;
53 if (sendmsg(sock, &msg, 0) < 0) {
61 static int recv_fd(int sock)
63 struct msghdr msg = {};
65 char buf[CMSG_SPACE(sizeof(int))] = {0}, c = 'c';
73 msg.msg_control = buf;
74 msg.msg_controllen = sizeof(buf);
76 if (recvmsg(sock, &msg, 0) < 0) {
81 cmsg = CMSG_FIRSTHDR(&msg);
83 return *((int *)CMSG_DATA(cmsg));
86 static int user_trap_syscall(int nr, unsigned int flags)
88 struct sock_filter filter[] = {
89 BPF_STMT(BPF_LD+BPF_W+BPF_ABS,
90 offsetof(struct seccomp_data, nr)),
91 BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, nr, 0, 1),
92 BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_USER_NOTIF),
93 BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW),
96 struct sock_fprog prog = {
97 .len = (unsigned short)ARRAY_SIZE(filter),
101 return seccomp(SECCOMP_SET_MODE_FILTER, flags, &prog);
104 static int handle_req(struct seccomp_notif *req,
105 struct seccomp_notif_resp *resp, int listener)
107 char path[PATH_MAX], source[PATH_MAX], target[PATH_MAX];
111 resp->error = -EPERM;
114 if (req->data.nr != __NR_mount) {
115 fprintf(stderr, "huh? trapped something besides mount? %d\n", req->data.nr);
119 /* Only allow bind mounts. */
120 if (!(req->data.args[3] & MS_BIND))
124 * Ok, let's read the task's memory to see where they wanted their
127 snprintf(path, sizeof(path), "/proc/%d/mem", req->pid);
128 mem = open(path, O_RDONLY);
135 * Now we avoid a TOCTOU: we referred to a pid by its pid, but since
136 * the pid that made the syscall may have died, we need to confirm that
137 * the pid is still valid after we open its /proc/pid/mem file. We can
138 * ask the listener fd this as follows.
140 * Note that this check should occur *after* any task-specific
141 * resources are opened, to make sure that the task has not died and
142 * we're not wrongly reading someone else's state in order to make
145 if (ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req->id) < 0) {
146 fprintf(stderr, "task died before we could map its memory\n");
151 * Phew, we've got the right /proc/pid/mem. Now we can read it. Note
152 * that to avoid another TOCTOU, we should read all of the pointer args
153 * before we decide to allow the syscall.
155 if (lseek(mem, req->data.args[0], SEEK_SET) < 0) {
160 ret = read(mem, source, sizeof(source));
166 if (lseek(mem, req->data.args[1], SEEK_SET) < 0) {
171 ret = read(mem, target, sizeof(target));
178 * Our policy is to only allow bind mounts inside /tmp. This isn't very
179 * interesting, because we could do unprivlieged bind mounts with user
180 * namespaces already, but you get the idea.
182 if (!strncmp(source, "/tmp/", 5) && !strncmp(target, "/tmp/", 5)) {
183 if (mount(source, target, NULL, req->data.args[3], NULL) < 0) {
185 perror("actual mount");
191 /* Even if we didn't allow it because of policy, generating the
192 * response was be a success, because we want to tell the worker EPERM.
203 int sk_pair[2], ret = 1, status, listener;
204 pid_t worker = 0 , tracer = 0;
206 if (socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair) < 0) {
207 perror("socketpair");
218 listener = user_trap_syscall(__NR_mount,
219 SECCOMP_FILTER_FLAG_NEW_LISTENER);
226 * Drop privileges. We definitely can't mount as uid 1000.
228 if (setuid(1000) < 0) {
234 * Send the listener to the parent; also serves as
237 if (send_fd(sk_pair[1], listener) < 0)
241 if (mkdir("/tmp/foo", 0755) < 0) {
247 * Try a bad mount just for grins.
249 if (mount("/dev/sda", "/tmp/foo", NULL, 0, NULL) != -1) {
250 fprintf(stderr, "huh? mounted /dev/sda?\n");
254 if (errno != EPERM) {
255 perror("bad error from mount");
260 * Ok, we expect this one to succeed.
262 if (mount("/tmp/foo", "/tmp/foo", NULL, MS_BIND, NULL) < 0) {
271 * Get the listener from the child.
273 listener = recv_fd(sk_pair[0]);
278 * Fork a task to handle the requests. This isn't strictly necessary,
279 * but it makes the particular writing of this sample easier, since we
280 * can just wait ofr the tracee to exit and kill the tracer.
289 struct seccomp_notif *req;
290 struct seccomp_notif_resp *resp;
291 struct seccomp_notif_sizes sizes;
293 if (seccomp(SECCOMP_GET_NOTIF_SIZES, 0, &sizes) < 0) {
294 perror("seccomp(GET_NOTIF_SIZES)");
298 req = malloc(sizes.seccomp_notif);
302 resp = malloc(sizes.seccomp_notif_resp);
305 memset(resp, 0, sizes.seccomp_notif_resp);
308 memset(req, 0, sizes.seccomp_notif);
309 if (ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, req)) {
310 perror("ioctl recv");
314 if (handle_req(req, resp, listener) < 0)
318 * ENOENT here means that the task may have gotten a
319 * signal and restarted the syscall. It's up to the
320 * handler to decide what to do in this case, but for
321 * the sample code, we just ignore it. Probably
322 * something better should happen, like undoing the
323 * mount, or keeping track of the args to make sure we
326 if (ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, resp) < 0 &&
328 perror("ioctl send");
343 if (waitpid(worker, &status, 0) != worker) {
348 if (umount2("/tmp/foo", MNT_DETACH) < 0 && errno != EINVAL) {
353 if (remove("/tmp/foo") < 0 && errno != ENOENT) {
358 if (!WIFEXITED(status) || WEXITSTATUS(status)) {
359 fprintf(stderr, "worker exited nonzero\n");
367 kill(tracer, SIGKILL);
369 kill(worker, SIGKILL);