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"
46 typedef enum MountMode {
47 /* This is ordered by priority! */
56 typedef struct BindMount {
63 static int append_mounts(BindMount **p, char **strv, MountMode mode) {
68 STRV_FOREACH(i, strv) {
72 if ((mode == INACCESSIBLE || mode == READONLY || mode == READWRITE) && (*i)[0] == '-') {
77 if (!path_is_absolute(*i))
88 static int mount_path_compare(const void *a, const void *b) {
89 const BindMount *p = a, *q = b;
91 if (path_equal(p->path, q->path)) {
93 /* If the paths are equal, check the mode */
94 if (p->mode < q->mode)
97 if (p->mode > q->mode)
103 /* If the paths are not equal, then order prefixes first */
104 if (path_startswith(p->path, q->path))
107 if (path_startswith(q->path, p->path))
113 static void drop_duplicates(BindMount *m, unsigned *n) {
114 BindMount *f, *t, *previous;
119 for (f = m, t = m, previous = NULL; f < m+*n; f++) {
121 /* The first one wins */
122 if (previous && path_equal(f->path, previous->path))
136 static int mount_dev(BindMount *m) {
137 static const char devnodes[] =
145 char temporary_mount[] = "/tmp/namespace-dev-XXXXXX";
146 const char *d, *dev = NULL, *devpts = NULL, *devshm = NULL, *devkdbus = NULL, *devhugepages = NULL, *devmqueue = NULL;
147 _cleanup_umask_ mode_t u;
154 if (!mkdtemp(temporary_mount))
157 dev = strappenda(temporary_mount, "/dev");
159 if (mount("tmpfs", dev, "tmpfs", MS_NOSUID|MS_STRICTATIME, "mode=755") < 0) {
164 devpts = strappenda(temporary_mount, "/dev/pts");
166 if (mount("/dev/pts", devpts, NULL, MS_BIND, NULL) < 0) {
171 devshm = strappenda(temporary_mount, "/dev/shm");
172 mkdir(devshm, 01777);
173 r = mount("/dev/shm", devshm, NULL, MS_BIND, NULL);
179 devmqueue = strappenda(temporary_mount, "/dev/mqueue");
180 mkdir(devmqueue, 0755);
181 mount("/dev/mqueue", devmqueue, NULL, MS_BIND, NULL);
183 devkdbus = strappenda(temporary_mount, "/dev/kdbus");
184 mkdir(devkdbus, 0755);
185 mount("/dev/kdbus", devkdbus, NULL, MS_BIND, NULL);
187 devhugepages = strappenda(temporary_mount, "/dev/hugepages");
188 mkdir(devhugepages, 0755);
189 mount("/dev/hugepages", devhugepages, NULL, MS_BIND, NULL);
191 NULSTR_FOREACH(d, devnodes) {
192 _cleanup_free_ char *dn = NULL;
205 if (!S_ISBLK(st.st_mode) &&
206 !S_ISCHR(st.st_mode)) {
214 dn = strappend(temporary_mount, d);
220 r = mknod(dn, st.st_mode, st.st_rdev);
227 dev_setup(temporary_mount);
229 if (mount(dev, "/dev/", NULL, MS_MOVE, NULL) < 0) {
235 rmdir(temporary_mount);
250 umount(devhugepages);
260 rmdir(temporary_mount);
265 static int apply_mount(
268 const char *var_tmp_dir) {
281 what = "/run/systemd/inaccessible";
293 case PRIVATE_VAR_TMP:
298 assert_not_reached("Unknown mode");
303 r = mount(what, m->path, NULL, MS_BIND|MS_REC, NULL);
305 log_debug("Successfully mounted %s to %s", what, m->path);
306 else if (m->ignore && errno == ENOENT)
312 static int make_read_only(BindMount *m) {
317 if (m->mode != INACCESSIBLE && m->mode != READONLY)
320 r = mount(NULL, m->path, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_REC, NULL);
321 if (r < 0 && !(m->ignore && errno == ENOENT))
328 char** read_write_dirs,
329 char** read_only_dirs,
330 char** inaccessible_dirs,
334 unsigned mount_flags) {
336 BindMount *m, *mounts = NULL;
340 if (mount_flags == 0)
341 mount_flags = MS_SHARED;
343 if (unshare(CLONE_NEWNS) < 0)
346 n = !!tmp_dir + !!var_tmp_dir +
347 strv_length(read_write_dirs) +
348 strv_length(read_only_dirs) +
349 strv_length(inaccessible_dirs) +
353 m = mounts = (BindMount *) alloca(n * sizeof(BindMount));
354 r = append_mounts(&m, read_write_dirs, READWRITE);
358 r = append_mounts(&m, read_only_dirs, READONLY);
362 r = append_mounts(&m, inaccessible_dirs, INACCESSIBLE);
368 m->mode = PRIVATE_TMP;
373 m->path = "/var/tmp";
374 m->mode = PRIVATE_VAR_TMP;
380 m->mode = PRIVATE_DEV;
384 assert(mounts + n == m);
386 qsort(mounts, n, sizeof(BindMount), mount_path_compare);
387 drop_duplicates(mounts, &n);
391 /* Remount / as SLAVE so that nothing now mounted in the namespace
392 shows up in the parent */
393 if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0)
396 for (m = mounts; m < mounts + n; ++m) {
397 r = apply_mount(m, tmp_dir, var_tmp_dir);
402 for (m = mounts; m < mounts + n; ++m) {
403 r = make_read_only(m);
409 /* Remount / as the desired mode. Not that this will not
410 * reestablish propagation from our side to the host, since
411 * what's disconnected is disconnected. */
412 if (mount(NULL, "/", NULL, mount_flags | MS_REC, NULL) < 0) {
421 for (m = mounts; m < mounts + n; ++m)
423 umount2(m->path, MNT_DETACH);
429 static int setup_one_tmp_dir(const char *id, const char *prefix, char **path) {
430 _cleanup_free_ char *x = NULL;
431 char bid[SD_ID128_STRING_MAX];
439 /* We include the boot id in the directory so that after a
440 * reboot we can easily identify obsolete directories. */
442 r = sd_id128_get_boot(&boot_id);
446 x = strjoin(prefix, "/systemd-private-", sd_id128_to_string(boot_id, bid), "-", id, "-XXXXXX", NULL);
454 RUN_WITH_UMASK(0000) {
457 y = strappenda(x, "/tmp");
459 if (mkdir(y, 0777 | S_ISVTX) < 0)
469 int setup_tmp_dirs(const char *id, char **tmp_dir, char **var_tmp_dir) {
477 r = setup_one_tmp_dir(id, "/tmp", &a);
481 r = setup_one_tmp_dir(id, "/var/tmp", &b);
485 t = strappenda(a, "/tmp");
499 int setup_netns(int netns_storage_socket[2]) {
500 _cleanup_close_ int netns = -1;
502 struct cmsghdr cmsghdr;
503 uint8_t buf[CMSG_SPACE(sizeof(int))];
506 .msg_control = &control,
507 .msg_controllen = sizeof(control),
509 struct cmsghdr *cmsg;
512 assert(netns_storage_socket);
513 assert(netns_storage_socket[0] >= 0);
514 assert(netns_storage_socket[1] >= 0);
516 /* We use the passed socketpair as a storage buffer for our
517 * namespace reference fd. Whatever process runs this first
518 * shall create a new namespace, all others should just join
519 * it. To serialize that we use a file lock on the socket
522 * It's a bit crazy, but hey, works great! */
524 if (lockf(netns_storage_socket[0], F_LOCK, 0) < 0)
527 if (recvmsg(netns_storage_socket[0], &mh, MSG_DONTWAIT|MSG_CMSG_CLOEXEC) < 0) {
528 if (errno != EAGAIN) {
533 /* Nothing stored yet, so let's create a new namespace */
535 if (unshare(CLONE_NEWNET) < 0) {
542 netns = open("/proc/self/ns/net", O_RDONLY|O_CLOEXEC|O_NOCTTY);
550 /* Yay, found something, so let's join the namespace */
552 for (cmsg = CMSG_FIRSTHDR(&mh); cmsg; cmsg = CMSG_NXTHDR(&mh, cmsg)) {
553 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
554 assert(cmsg->cmsg_len == CMSG_LEN(sizeof(int)));
555 netns = *(int*) CMSG_DATA(cmsg);
559 if (setns(netns, CLONE_NEWNET) < 0) {
567 cmsg = CMSG_FIRSTHDR(&mh);
568 cmsg->cmsg_level = SOL_SOCKET;
569 cmsg->cmsg_type = SCM_RIGHTS;
570 cmsg->cmsg_len = CMSG_LEN(sizeof(int));
571 memcpy(CMSG_DATA(cmsg), &netns, sizeof(int));
572 mh.msg_controllen = cmsg->cmsg_len;
574 if (sendmsg(netns_storage_socket[1], &mh, MSG_DONTWAIT|MSG_NOSIGNAL) < 0) {
580 lockf(netns_storage_socket[0], F_ULOCK, 0);