1 /* vi: set sw=4 ts=4: */
3 * Mini unshare implementation for busybox.
5 * Copyright (C) 2016 by Bartosz Golaszewski <bartekgola@gmail.com>
7 * Licensed under GPLv2 or later, see file LICENSE in this source tree.
10 //config:config UNSHARE
11 //config: bool "unshare"
13 //config: depends on LONG_OPTS && !NOMMU
14 //config: select PLATFORM_LINUX
16 //config: Run program with some namespaces unshared from parent.
18 // depends on LONG_OPTS: it is awkward to exclude code which handles --propagation
19 // and --setgroups based on LONG_OPTS, so instead applet requires LONG_OPTS.
20 // depends on !NOMMU: we need fork()
22 //applet:IF_UNSHARE(APPLET(unshare, BB_DIR_USR_BIN, BB_SUID_DROP))
24 //kbuild:lib-$(CONFIG_UNSHARE) += unshare.o
26 //usage:#define unshare_trivial_usage
27 //usage: "[OPTIONS] [PROG [ARGS]]"
28 //usage:#define unshare_full_usage "\n"
29 //usage: "\n -m, --mount[=FILE] Unshare mount namespace"
30 //usage: "\n -u, --uts[=FILE] Unshare UTS namespace (hostname etc.)"
31 //usage: "\n -i, --ipc[=FILE] Unshare System V IPC namespace"
32 //usage: "\n -n, --net[=FILE] Unshare network namespace"
33 //usage: "\n -p, --pid[=FILE] Unshare PID namespace"
34 //usage: "\n -U, --user[=FILE} Unshare user namespace"
35 //usage: "\n -f, --fork Fork before execing PROG"
36 //usage: "\n -r, --map-root-user Map current user to root (implies -u)"
37 //usage: "\n --mount-proc[=DIR] Mount /proc filesystem first (implies -m)"
38 //usage: "\n --propagation slave|shared|private|unchanged"
39 //usage: "\n Modify mount propagation in mount namespace"
40 //usage: "\n --setgroups allow|deny Control the setgroups syscall in user namespaces"
43 #include <sys/mount.h>
46 static void mount_or_die(const char *source, const char *target,
47 const char *fstype, unsigned long mountflags)
49 if (mount(source, target, fstype, mountflags, NULL)) {
50 bb_perror_msg_and_die("can't mount %s on %s (flags:0x%lx)",
51 source, target, mountflags);
52 /* fstype is always either NULL or "proc".
53 * "proc" is only used to mount /proc.
54 * No need to clutter up error message with fstype,
55 * it is easily deductible.
60 // TODO: move to libbb
61 static int wait_for_exitstatus(pid_t pid)
65 n = safe_waitpid(pid, &exit_status, 0);
67 bb_perror_msg_and_die("waitpid");
72 * Longest possible path to a procfs file used in unshare. Must be able to
73 * contain the '/proc/' string, the '/ns/user' string which is the longest
74 * namespace name and a 32-bit integer representing the process ID.
76 #define PATH_PROC_SETGROUPS "/proc/self/setgroups"
77 #define PATH_PROC_UIDMAP "/proc/self/uid_map"
78 #define PATH_PROC_GIDMAP "/proc/self/gid_map"
80 struct namespace_descr {
82 const char nsfile4[4];
85 struct namespace_ctx {
95 OPT_user = 1 << 5, /* OPT_user, NS_USR_POS, and ns_list[] index must match! */
97 OPT_map_root = 1 << 7,
98 OPT_mount_proc = 1 << 8,
99 OPT_propagation = 1 << 9,
100 OPT_setgroups = 1 << 10,
108 NS_USR_POS, /* OPT_user, NS_USR_POS, and ns_list[] index must match! */
111 static const struct namespace_descr ns_list[] = {
112 { CLONE_NEWNS, "mnt" },
113 { CLONE_NEWUTS, "uts" },
114 { CLONE_NEWIPC, "ipc" },
115 { CLONE_NEWNET, "net" },
116 { CLONE_NEWPID, "pid" },
117 { CLONE_NEWUSER, "user" }, /* OPT_user, NS_USR_POS, and ns_list[] index must match! */
121 * Upstream unshare doesn't support short options for --mount-proc,
122 * --propagation, --setgroups.
123 * Optional arguments (namespace mountpoints) exist only for long opts,
124 * we are forced to use "fake" letters for them.
125 * '+': stop at first non-option.
127 static const char opt_str[] = "+muinpU""fr""\xfd::""\xfe:""\xff:";
128 static const char unshare_longopts[] ALIGN1 =
129 "mount\0" Optional_argument "\xf0"
130 "uts\0" Optional_argument "\xf1"
131 "ipc\0" Optional_argument "\xf2"
132 "network\0" Optional_argument "\xf3"
133 "pid\0" Optional_argument "\xf4"
134 "user\0" Optional_argument "\xf5"
135 "fork\0" No_argument "f"
136 "map-root-user\0" No_argument "r"
137 "mount-proc\0" Optional_argument "\xfd"
138 "propagation\0" Required_argument "\xfe"
139 "setgroups\0" Required_argument "\xff"
142 /* Ugly-looking string reuse trick */
143 #define PRIVATE_STR "private\0""unchanged\0""shared\0""slave\0"
144 #define PRIVATE_UNCHANGED_SHARED_SLAVE PRIVATE_STR
146 static unsigned long parse_propagation(const char *prop_str)
148 int i = index_in_strings(PRIVATE_UNCHANGED_SHARED_SLAVE, prop_str);
150 bb_error_msg_and_die("unrecognized: --%s=%s", "propagation", prop_str);
152 return MS_REC | MS_PRIVATE;
156 return MS_REC | MS_SHARED;
157 return MS_REC | MS_SLAVE;
160 static void mount_namespaces(pid_t pid, struct namespace_ctx *ns_ctx_list)
162 const struct namespace_descr *ns;
163 struct namespace_ctx *ns_ctx;
166 for (i = 0; i < NS_COUNT; i++) {
167 char nsf[sizeof("/proc/%u/ns/AAAA") + sizeof(int)*3];
170 ns_ctx = &ns_ctx_list[i];
173 sprintf(nsf, "/proc/%u/ns/%.4s", (unsigned)pid, ns->nsfile4);
174 mount_or_die(nsf, ns_ctx->path, NULL, MS_BIND);
178 int unshare_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
179 int unshare_main(int argc UNUSED_PARAM, char **argv)
184 uintptr_t need_mount;
185 const char *proc_mnt_target;
186 const char *prop_str;
187 const char *setgrp_str;
188 unsigned long prop_flags;
189 uid_t reuid = geteuid();
190 gid_t regid = getegid();
192 pid_t child = child; /* for compiler */
193 struct namespace_ctx ns_ctx_list[NS_COUNT];
195 memset(ns_ctx_list, 0, sizeof(ns_ctx_list));
196 proc_mnt_target = "/proc";
197 prop_str = PRIVATE_STR;
201 "\xf0""m" /* long opts (via their "fake chars") imply short opts */
207 ":ru" /* --map-root-user or -r implies -u */
208 ":\xfd""m" /* --mount-proc implies -m */
210 applet_long_options = unshare_longopts;
211 opts = getopt32(argv, opt_str,
212 &proc_mnt_target, &prop_str, &setgrp_str,
213 &ns_ctx_list[NS_MNT_POS].path,
214 &ns_ctx_list[NS_UTS_POS].path,
215 &ns_ctx_list[NS_IPC_POS].path,
216 &ns_ctx_list[NS_NET_POS].path,
217 &ns_ctx_list[NS_PID_POS].path,
218 &ns_ctx_list[NS_USR_POS].path
221 //bb_error_msg("opts:0x%x", opts);
222 //bb_error_msg("mount:%s", ns_ctx_list[NS_MNT_POS].path);
223 //bb_error_msg("proc_mnt_target:%s", proc_mnt_target);
224 //bb_error_msg("prop_str:%s", prop_str);
225 //bb_error_msg("setgrp_str:%s", setgrp_str);
229 if (strcmp(setgrp_str, "allow") == 0) {
230 if (opts & OPT_map_root) {
231 bb_error_msg_and_die(
232 "--setgroups=allow and --map-root-user "
233 "are mutually exclusive"
237 /* It's not "allow", must be "deny" */
238 if (strcmp(setgrp_str, "deny") != 0)
239 bb_error_msg_and_die("unrecognized: --%s=%s",
240 "setgroups", setgrp_str);
246 for (i = 0; i < NS_COUNT; i++) {
247 const struct namespace_descr *ns = &ns_list[i];
248 struct namespace_ctx *ns_ctx = &ns_ctx_list[i];
251 unsflags |= ns->flag;
253 need_mount |= (uintptr_t)(ns_ctx->path);
255 /* need_mount != 0 if at least one FILE was given */
257 prop_flags = MS_REC | MS_PRIVATE;
258 /* Silently ignore --propagation if --mount is not requested. */
259 if (opts & OPT_mount)
260 prop_flags = parse_propagation(prop_str);
263 * Special case: if we were requested to unshare the mount namespace
264 * AND to make any namespace persistent (by bind mounting it) we need
265 * to spawn a child process which will wait for the parent to call
266 * unshare(), then mount parent's namespaces while still in the
267 * previous namespace.
270 if (need_mount && (opts & OPT_mount)) {
272 * Can't use getppid() in child, as we can be unsharing the
275 pid_t ppid = getpid();
284 /* Wait until parent calls unshare() */
285 read(fdp.rd, ns_ctx_list, 1); /* ...using bogus buffer */
288 /* Mount parent's unshared namespaces. */
289 mount_namespaces(ppid, ns_ctx_list);
292 /* Parent continues */
295 if (unshare(unsflags) != 0)
296 bb_perror_msg_and_die("unshare(0x%x)", unsflags);
299 close(fdp.wr); /* Release child */
304 /* Wait for the child to finish mounting the namespaces. */
305 if (opts & OPT_mount) {
306 int exit_status = wait_for_exitstatus(child);
307 if (WIFEXITED(exit_status) &&
308 WEXITSTATUS(exit_status) != EXIT_SUCCESS)
309 return WEXITSTATUS(exit_status);
312 * Regular way - we were requested to mount some other
313 * namespaces: mount them after the call to unshare().
315 mount_namespaces(getpid(), ns_ctx_list);
320 * When we're unsharing the pid namespace, it's not the process that
321 * calls unshare() that is put into the new namespace, but its first
322 * child. The user may want to use this option to spawn a new process
323 * that'll become PID 1 in this new namespace.
325 if (opts & OPT_fork) {
329 int exit_status = wait_for_exitstatus(pid);
330 if (WIFSIGNALED(exit_status))
331 kill_myself_with_sig(WTERMSIG(exit_status));
332 return WEXITSTATUS(exit_status);
334 /* Child continues */
337 if (opts & OPT_map_root) {
338 char uidmap_buf[sizeof("%u 0 1") + sizeof(int)*3];
341 * Since Linux 3.19 unprivileged writing of /proc/self/gid_map
342 * has been disabled unless /proc/self/setgroups is written
343 * first to permanently disable the ability to call setgroups
344 * in that user namespace.
346 xopen_xwrite_close(PATH_PROC_SETGROUPS, "deny");
347 sprintf(uidmap_buf, "%u 0 1", (unsigned)reuid);
348 xopen_xwrite_close(PATH_PROC_UIDMAP, uidmap_buf);
349 sprintf(uidmap_buf, "%u 0 1", (unsigned)regid);
350 xopen_xwrite_close(PATH_PROC_GIDMAP, uidmap_buf);
353 /* Write "allow" or "deny" */
354 xopen_xwrite_close(PATH_PROC_SETGROUPS, setgrp_str);
357 if (opts & OPT_mount) {
358 mount_or_die("none", "/", NULL, prop_flags);
361 if (opts & OPT_mount_proc) {
363 * When creating a new pid namespace, we might want the pid
364 * subdirectories in /proc to remain consistent with the new
365 * process IDs. Without --mount-proc the pids in /proc would
366 * still reflect the old pid namespace. This is why we make
367 * /proc private here and then do a fresh mount.
369 mount_or_die("none", proc_mnt_target, NULL, MS_PRIVATE | MS_REC);
370 mount_or_die("proc", proc_mnt_target, "proc", MS_NOSUID | MS_NOEXEC | MS_NODEV);
374 BB_EXECVP_or_die(argv);
376 /* unshare from util-linux 2.27.1, despite not documenting it,
377 * runs a login shell (argv0="-sh") if no PROG is given
379 run_shell(getenv("SHELL"), /*login:*/ 1, NULL, NULL);