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 General Public License as published by
10 the Free Software Foundation; either version 2 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 General Public License for more details.
18 You should have received a copy of the GNU 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>
36 #include "namespace.h"
39 typedef enum PathMode {
40 /* This is ordered by priority! */
52 static int append_paths(Path **p, char **strv, PathMode mode) {
55 STRV_FOREACH(i, strv) {
57 if (!path_is_absolute(*i))
68 static int path_compare(const void *a, const void *b) {
69 const Path *p = a, *q = b;
71 if (path_equal(p->path, q->path)) {
73 /* If the paths are equal, check the mode */
74 if (p->mode < q->mode)
77 if (p->mode > q->mode)
83 /* If the paths are not equal, then order prefixes first */
84 if (path_startswith(p->path, q->path))
87 if (path_startswith(q->path, p->path))
93 static void drop_duplicates(Path *p, unsigned *n, bool *need_inaccessible, bool *need_private) {
94 Path *f, *t, *previous;
98 assert(need_inaccessible);
101 for (f = p, t = p, previous = NULL; f < p+*n; f++) {
103 if (previous && path_equal(f->path, previous->path))
109 if (t->mode == PRIVATE)
110 *need_private = true;
112 if (t->mode == INACCESSIBLE)
113 *need_inaccessible = true;
123 static int apply_mount(Path *p, const char *root_dir, const char *inaccessible_dir, const char *private_dir, unsigned long flags) {
130 assert(inaccessible_dir);
133 if (!(where = strappend(root_dir, p->path)))
139 what = inaccessible_dir;
156 assert_not_reached("Unknown mode");
159 if ((r = mount(what, where, NULL, MS_BIND|MS_REC, NULL)) >= 0) {
160 log_debug("Successfully mounted %s to %s", what, where);
162 /* The bind mount will always inherit the original
163 * flags. If we want to set any flag we need
164 * to do so in a second indepdant step. */
166 r = mount(NULL, where, NULL, MS_REMOUNT|MS_BIND|MS_REC|flags, NULL);
168 /* Avoid expontial growth of trees */
169 if (r >= 0 && path_equal(p->path, "/"))
170 r = mount(NULL, where, NULL, MS_REMOUNT|MS_BIND|MS_UNBINDABLE|flags, NULL);
174 umount2(where, MNT_DETACH);
187 unsigned long flags) {
190 tmp_dir[] = "/tmp/systemd-namespace-XXXXXX",
191 root_dir[] = "/tmp/systemd-namespace-XXXXXX/root",
192 old_root_dir[] = "/tmp/systemd-namespace-XXXXXX/root/tmp/old-root-XXXXXX",
193 inaccessible_dir[] = "/tmp/systemd-namespace-XXXXXX/inaccessible",
194 private_dir[] = "/tmp/systemd-namespace-XXXXXX/private";
198 bool need_private = false, need_inaccessible = false;
199 bool remove_tmp = false, remove_root = false, remove_old_root = false, remove_inaccessible = false, remove_private = false;
204 strv_length(writable) +
205 strv_length(readable) +
206 strv_length(inaccessible) +
207 (private_tmp ? 2 : 1);
209 if (!(paths = new(Path, n)))
213 if ((r = append_paths(&p, writable, READWRITE)) < 0 ||
214 (r = append_paths(&p, readable, READONLY)) < 0 ||
215 (r = append_paths(&p, inaccessible, INACCESSIBLE)) < 0)
228 assert(paths + n == p);
230 qsort(paths, n, sizeof(Path), path_compare);
231 drop_duplicates(paths, &n, &need_inaccessible, &need_private);
233 if (!mkdtemp(tmp_dir)) {
239 memcpy(root_dir, tmp_dir, sizeof(tmp_dir)-1);
240 if (mkdir(root_dir, 0777) < 0) {
246 if (need_inaccessible) {
247 memcpy(inaccessible_dir, tmp_dir, sizeof(tmp_dir)-1);
248 if (mkdir(inaccessible_dir, 0) < 0) {
252 remove_inaccessible = true;
256 memcpy(private_dir, tmp_dir, sizeof(tmp_dir)-1);
257 if (mkdir(private_dir, 0777 + S_ISVTX) < 0) {
261 remove_private = true;
264 if (unshare(CLONE_NEWNS) < 0) {
269 /* We assume that by default mount events from us won't be
270 * propagated to the root namespace. */
272 for (p = paths; p < paths + n; p++)
273 if ((r = apply_mount(p, root_dir, inaccessible_dir, private_dir, flags)) < 0)
276 memcpy(old_root_dir, tmp_dir, sizeof(tmp_dir)-1);
277 if (!mkdtemp(old_root_dir)) {
281 remove_old_root = true;
283 if (chdir(root_dir) < 0) {
288 if (pivot_root(root_dir, old_root_dir) < 0) {
293 t = old_root_dir + sizeof(root_dir) - 1;
294 if (umount2(t, MNT_DETACH) < 0)
295 /* At this point it's too late to turn anything back,
296 * since we are already in the new root. */
306 for (p--; p >= paths; p--) {
307 char full_path[PATH_MAX];
309 snprintf(full_path, sizeof(full_path), "%s%s", root_dir, p->path);
310 char_array_0(full_path);
312 umount2(full_path, MNT_DETACH);
319 if (remove_inaccessible)
320 rmdir(inaccessible_dir);