1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
4 This file is part of systemd.
6 Copyright 2010 Lennart Poettering
8 systemd is free software; you can redistribute it and/or modify it
9 under the terms of the GNU Lesser General Public License as published by
10 the Free Software Foundation; either version 2.1 of the License, or
11 (at your option) any later version.
13 systemd is distributed in the hope that it will be useful, but
14 WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 Lesser General Public License for more details.
18 You should have received a copy of the GNU Lesser General Public License
19 along with systemd; If not, see <http://www.gnu.org/licenses/>.
23 #include <sys/mount.h>
28 #include <sys/types.h>
30 #include <sys/syscall.h>
37 #include "path-util.h"
38 #include "namespace.h"
41 #include "loopback-setup.h"
43 #include "dev-setup.h"
47 typedef enum MountMode {
48 /* This is ordered by priority! */
58 typedef struct BindMount {
65 static int append_mounts(BindMount **p, char **strv, MountMode mode) {
70 STRV_FOREACH(i, strv) {
75 if ((mode == INACCESSIBLE || mode == READONLY || mode == READWRITE) && (*i)[0] == '-') {
80 if (!path_is_absolute(*i))
91 static int mount_path_compare(const void *a, const void *b) {
92 const BindMount *p = a, *q = b;
94 if (path_equal(p->path, q->path)) {
96 /* If the paths are equal, check the mode */
97 if (p->mode < q->mode)
100 if (p->mode > q->mode)
106 /* If the paths are not equal, then order prefixes first */
107 if (path_startswith(p->path, q->path))
110 if (path_startswith(q->path, p->path))
116 static void drop_duplicates(BindMount *m, unsigned *n) {
117 BindMount *f, *t, *previous;
122 for (f = m, t = m, previous = NULL; f < m+*n; f++) {
124 /* The first one wins */
125 if (previous && path_equal(f->path, previous->path))
138 static int mount_dev(BindMount *m) {
139 static const char devnodes[] =
147 char temporary_mount[] = "/tmp/namespace-dev-XXXXXX";
148 const char *d, *dev = NULL, *devpts = NULL, *devshm = NULL, *devkdbus = NULL, *devhugepages = NULL, *devmqueue = NULL, *devlog = NULL, *devptmx = NULL;
149 _cleanup_umask_ mode_t u;
156 if (!mkdtemp(temporary_mount))
159 dev = strappenda(temporary_mount, "/dev");
161 if (mount("tmpfs", dev, "tmpfs", MS_NOSUID|MS_STRICTATIME, "mode=755") < 0) {
166 devpts = strappenda(temporary_mount, "/dev/pts");
168 if (mount("/dev/pts", devpts, NULL, MS_BIND, NULL) < 0) {
173 devptmx = strappenda(temporary_mount, "/dev/ptmx");
174 symlink("pts/ptmx", devptmx);
176 devshm = strappenda(temporary_mount, "/dev/shm");
177 mkdir(devshm, 01777);
178 r = mount("/dev/shm", devshm, NULL, MS_BIND, NULL);
184 devmqueue = strappenda(temporary_mount, "/dev/mqueue");
185 mkdir(devmqueue, 0755);
186 mount("/dev/mqueue", devmqueue, NULL, MS_BIND, NULL);
188 devkdbus = strappenda(temporary_mount, "/dev/kdbus");
189 mkdir(devkdbus, 0755);
190 mount("/dev/kdbus", devkdbus, NULL, MS_BIND, NULL);
192 devhugepages = strappenda(temporary_mount, "/dev/hugepages");
193 mkdir(devhugepages, 0755);
194 mount("/dev/hugepages", devhugepages, NULL, MS_BIND, NULL);
196 devlog = strappenda(temporary_mount, "/dev/log");
197 symlink("/run/systemd/journal/dev-log", devlog);
199 NULSTR_FOREACH(d, devnodes) {
200 _cleanup_free_ char *dn = NULL;
213 if (!S_ISBLK(st.st_mode) &&
214 !S_ISCHR(st.st_mode)) {
222 dn = strappend(temporary_mount, d);
228 label_context_set(d, st.st_mode);
229 r = mknod(dn, st.st_mode, st.st_rdev);
230 label_context_clear();
238 dev_setup(temporary_mount);
240 if (mount(dev, "/dev/", NULL, MS_MOVE, NULL) < 0) {
246 rmdir(temporary_mount);
261 umount(devhugepages);
271 rmdir(temporary_mount);
276 static int mount_kdbus(BindMount *m) {
278 char temporary_mount[] = "/tmp/kdbus-dev-XXXXXX";
279 _cleanup_free_ char *basepath = NULL;
280 _cleanup_umask_ mode_t u;
281 char *busnode = NULL, *root;
289 if (!mkdtemp(temporary_mount)) {
290 log_error("Failed create temp dir: %m");
294 root = strappenda(temporary_mount, "/kdbus");
296 if (mount("tmpfs", root, "tmpfs", MS_NOSUID|MS_STRICTATIME, "mode=777") < 0) {
301 /* create a new /dev/null dev node copy so we have some fodder to
302 * bind-mount the custom endpoint over. */
303 if (stat("/dev/null", &st) < 0) {
304 log_error("Failed to stat /dev/null: %m");
309 busnode = strappenda(root, "/bus");
310 if (mknod(busnode, (st.st_mode & ~07777) | 0600, st.st_rdev) < 0) {
311 log_error("mknod() for %s failed: %m", busnode);
316 r = mount(m->path, busnode, "bind", MS_BIND, NULL);
318 log_error("bind mount of %s failed: %m", m->path);
323 basepath = dirname_malloc(m->path);
329 if (mount(root, basepath, NULL, MS_MOVE, NULL) < 0) {
330 log_error("bind mount of %s failed: %m", basepath);
335 rmdir(temporary_mount);
349 rmdir(temporary_mount);
354 static int apply_mount(
357 const char *var_tmp_dir) {
368 /* First, get rid of everything that is below if there
369 * is anything... Then, overmount it with an
370 * inaccessible directory. */
371 umount_recursive(m->path, 0);
373 what = "/run/systemd/inaccessible";
378 /* Nothing to mount here, we just later toggle the
379 * MS_RDONLY bit for the mount point */
386 case PRIVATE_VAR_TMP:
393 case PRIVATE_BUS_ENDPOINT:
394 return mount_kdbus(m);
397 assert_not_reached("Unknown mode");
402 r = mount(what, m->path, NULL, MS_BIND|MS_REC, NULL);
404 log_debug("Successfully mounted %s to %s", what, m->path);
405 else if (m->ignore && errno == ENOENT)
411 static int make_read_only(BindMount *m) {
416 if (IN_SET(m->mode, INACCESSIBLE, READONLY))
417 r = bind_remount_recursive(m->path, true);
418 else if (IN_SET(m->mode, READWRITE, PRIVATE_TMP, PRIVATE_VAR_TMP, PRIVATE_DEV))
419 r = bind_remount_recursive(m->path, false);
423 if (m->ignore && r == -ENOENT)
430 char** read_write_dirs,
431 char** read_only_dirs,
432 char** inaccessible_dirs,
435 char* bus_endpoint_path,
437 ProtectHome protect_home,
438 ProtectSystem protect_system,
439 unsigned mount_flags) {
441 BindMount *m, *mounts = NULL;
445 if (mount_flags == 0)
446 mount_flags = MS_SHARED;
448 if (unshare(CLONE_NEWNS) < 0)
451 n = !!tmp_dir + !!var_tmp_dir + !!bus_endpoint_path +
452 strv_length(read_write_dirs) +
453 strv_length(read_only_dirs) +
454 strv_length(inaccessible_dirs) +
456 (protect_home != PROTECT_HOME_NO ? 3 : 0) +
457 (protect_system != PROTECT_SYSTEM_NO ? 2 : 0) +
458 (protect_system == PROTECT_SYSTEM_FULL ? 1 : 0);
461 m = mounts = (BindMount *) alloca0(n * sizeof(BindMount));
462 r = append_mounts(&m, read_write_dirs, READWRITE);
466 r = append_mounts(&m, read_only_dirs, READONLY);
470 r = append_mounts(&m, inaccessible_dirs, INACCESSIBLE);
476 m->mode = PRIVATE_TMP;
481 m->path = "/var/tmp";
482 m->mode = PRIVATE_VAR_TMP;
488 m->mode = PRIVATE_DEV;
492 if (bus_endpoint_path) {
493 m->path = bus_endpoint_path;
494 m->mode = PRIVATE_BUS_ENDPOINT;
498 if (protect_home != PROTECT_HOME_NO) {
499 r = append_mounts(&m, STRV_MAKE("-/home", "-/run/user", "-/root"), protect_home == PROTECT_HOME_READ_ONLY ? READONLY : INACCESSIBLE);
504 if (protect_system != PROTECT_SYSTEM_NO) {
505 r = append_mounts(&m, protect_system == PROTECT_SYSTEM_FULL ? STRV_MAKE("/usr", "-/boot", "/etc") : STRV_MAKE("/usr", "-/boot"), READONLY);
510 assert(mounts + n == m);
512 qsort(mounts, n, sizeof(BindMount), mount_path_compare);
513 drop_duplicates(mounts, &n);
517 /* Remount / as SLAVE so that nothing now mounted in the namespace
518 shows up in the parent */
519 if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0)
522 for (m = mounts; m < mounts + n; ++m) {
523 r = apply_mount(m, tmp_dir, var_tmp_dir);
528 for (m = mounts; m < mounts + n; ++m) {
529 r = make_read_only(m);
535 /* Remount / as the desired mode. Not that this will not
536 * reestablish propagation from our side to the host, since
537 * what's disconnected is disconnected. */
538 if (mount(NULL, "/", NULL, mount_flags | MS_REC, NULL) < 0) {
547 for (m = mounts; m < mounts + n; ++m)
549 umount2(m->path, MNT_DETACH);
555 static int setup_one_tmp_dir(const char *id, const char *prefix, char **path) {
556 _cleanup_free_ char *x = NULL;
557 char bid[SD_ID128_STRING_MAX];
565 /* We include the boot id in the directory so that after a
566 * reboot we can easily identify obsolete directories. */
568 r = sd_id128_get_boot(&boot_id);
572 x = strjoin(prefix, "/systemd-private-", sd_id128_to_string(boot_id, bid), "-", id, "-XXXXXX", NULL);
580 RUN_WITH_UMASK(0000) {
583 y = strappenda(x, "/tmp");
585 if (mkdir(y, 0777 | S_ISVTX) < 0)
595 int setup_tmp_dirs(const char *id, char **tmp_dir, char **var_tmp_dir) {
603 r = setup_one_tmp_dir(id, "/tmp", &a);
607 r = setup_one_tmp_dir(id, "/var/tmp", &b);
611 t = strappenda(a, "/tmp");
625 int setup_netns(int netns_storage_socket[2]) {
626 _cleanup_close_ int netns = -1;
628 struct cmsghdr cmsghdr;
629 uint8_t buf[CMSG_SPACE(sizeof(int))];
632 .msg_control = &control,
633 .msg_controllen = sizeof(control),
635 struct cmsghdr *cmsg;
638 assert(netns_storage_socket);
639 assert(netns_storage_socket[0] >= 0);
640 assert(netns_storage_socket[1] >= 0);
642 /* We use the passed socketpair as a storage buffer for our
643 * namespace reference fd. Whatever process runs this first
644 * shall create a new namespace, all others should just join
645 * it. To serialize that we use a file lock on the socket
648 * It's a bit crazy, but hey, works great! */
650 if (lockf(netns_storage_socket[0], F_LOCK, 0) < 0)
653 if (recvmsg(netns_storage_socket[0], &mh, MSG_DONTWAIT|MSG_CMSG_CLOEXEC) < 0) {
654 if (errno != EAGAIN) {
659 /* Nothing stored yet, so let's create a new namespace */
661 if (unshare(CLONE_NEWNET) < 0) {
668 netns = open("/proc/self/ns/net", O_RDONLY|O_CLOEXEC|O_NOCTTY);
676 /* Yay, found something, so let's join the namespace */
678 for (cmsg = CMSG_FIRSTHDR(&mh); cmsg; cmsg = CMSG_NXTHDR(&mh, cmsg)) {
679 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
680 assert(cmsg->cmsg_len == CMSG_LEN(sizeof(int)));
681 netns = *(int*) CMSG_DATA(cmsg);
685 if (setns(netns, CLONE_NEWNET) < 0) {
693 cmsg = CMSG_FIRSTHDR(&mh);
694 cmsg->cmsg_level = SOL_SOCKET;
695 cmsg->cmsg_type = SCM_RIGHTS;
696 cmsg->cmsg_len = CMSG_LEN(sizeof(int));
697 memcpy(CMSG_DATA(cmsg), &netns, sizeof(int));
698 mh.msg_controllen = cmsg->cmsg_len;
700 if (sendmsg(netns_storage_socket[1], &mh, MSG_DONTWAIT|MSG_NOSIGNAL) < 0) {
706 lockf(netns_storage_socket[0], F_ULOCK, 0);
711 static const char *const protect_home_table[_PROTECT_HOME_MAX] = {
712 [PROTECT_HOME_NO] = "no",
713 [PROTECT_HOME_YES] = "yes",
714 [PROTECT_HOME_READ_ONLY] = "read-only",
717 DEFINE_STRING_TABLE_LOOKUP(protect_home, ProtectHome);
719 static const char *const protect_system_table[_PROTECT_SYSTEM_MAX] = {
720 [PROTECT_SYSTEM_NO] = "no",
721 [PROTECT_SYSTEM_YES] = "yes",
722 [PROTECT_SYSTEM_FULL] = "full",
725 DEFINE_STRING_TABLE_LOOKUP(protect_system, ProtectSystem);