]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/mount-setup.c
cgroup: be nice to Ingo Molnar
[thirdparty/systemd.git] / src / mount-setup.c
1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
2
3 /***
4 This file is part of systemd.
5
6 Copyright 2010 Lennart Poettering
7
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.
12
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.
17
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/>.
20 ***/
21
22 #include <sys/mount.h>
23 #include <errno.h>
24 #include <sys/stat.h>
25 #include <stdlib.h>
26 #include <string.h>
27 #include <libgen.h>
28 #include <assert.h>
29 #include <unistd.h>
30 #include <ftw.h>
31
32 #include "mount-setup.h"
33 #include "log.h"
34 #include "macro.h"
35 #include "util.h"
36 #include "label.h"
37
38 #ifndef TTY_GID
39 #define TTY_GID 5
40 #endif
41
42 typedef struct MountPoint {
43 const char *what;
44 const char *where;
45 const char *type;
46 const char *options;
47 unsigned long flags;
48 bool fatal;
49 } MountPoint;
50
51 static const MountPoint mount_table[] = {
52 { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, true },
53 { "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, true },
54 { "devtmpfs", "/dev", "devtmpfs", "mode=755", MS_NOSUID, true },
55 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV, true },
56 { "devpts", "/dev/pts", "devpts", "mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC, false },
57 { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID|MS_NOEXEC|MS_NODEV, true },
58 { "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=755", MS_NOSUID|MS_NOEXEC|MS_NODEV, false },
59 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV, false },
60 };
61
62 /* These are API file systems that might be mounted by other software,
63 * we just list them here so that we know that we should ignore them */
64
65 static const char * const ignore_paths[] = {
66 "/selinux",
67 "/proc/bus/usb",
68 "/var/lib/nfs/rpc_pipefs",
69 "/proc/fs/nfsd"
70 };
71
72 bool mount_point_is_api(const char *path) {
73 unsigned i;
74
75 /* Checks if this mount point is considered "API", and hence
76 * should be ignored */
77
78 for (i = 0; i < ELEMENTSOF(mount_table); i ++)
79 if (path_equal(path, mount_table[i].where))
80 return true;
81
82 return path_startswith(path, "/sys/fs/cgroup/");
83 }
84
85 bool mount_point_ignore(const char *path) {
86 unsigned i;
87
88 for (i = 0; i < ELEMENTSOF(ignore_paths); i++)
89 if (path_equal(path, ignore_paths[i]))
90 return true;
91
92 return false;
93 }
94
95 static int mount_one(const MountPoint *p) {
96 int r;
97
98 assert(p);
99
100 /* Relabel first, just in case */
101 label_fix(p->where, true);
102
103 if ((r = path_is_mount_point(p->where)) < 0)
104 return r;
105
106 if (r > 0)
107 return 0;
108
109 /* The access mode here doesn't really matter too much, since
110 * the mounted file system will take precedence anyway. */
111 mkdir_p(p->where, 0755);
112
113 log_debug("Mounting %s to %s of type %s with options %s.",
114 p->what,
115 p->where,
116 p->type,
117 strna(p->options));
118
119 if (mount(p->what,
120 p->where,
121 p->type,
122 p->flags,
123 p->options) < 0) {
124 log_error("Failed to mount %s: %s", p->where, strerror(errno));
125 return p->fatal ? -errno : 0;
126 }
127
128 /* Relabel again, since we now mounted something fresh here */
129 label_fix(p->where, false);
130
131 return 0;
132 }
133
134 static int mount_cgroup_controllers(void) {
135 int r;
136 FILE *f;
137 char buf[LINE_MAX];
138
139 /* Mount all available cgroup controllers that are built into the kernel. */
140
141 if (!(f = fopen("/proc/cgroups", "re")))
142 return -ENOENT;
143
144 /* Ignore the header line */
145 (void) fgets(buf, sizeof(buf), f);
146
147 for (;;) {
148 MountPoint p;
149 char *controller, *where;
150 int enabled = false;
151
152 if (fscanf(f, "%ms %*i %*i %i", &controller, &enabled) != 2) {
153
154 if (feof(f))
155 break;
156
157 log_error("Failed to parse /proc/cgroups.");
158 r = -EIO;
159 goto finish;
160 }
161
162 if (!enabled) {
163 free(controller);
164 continue;
165 }
166
167 if (asprintf(&where, "/sys/fs/cgroup/%s", controller) < 0) {
168 free(controller);
169 r = -ENOMEM;
170 goto finish;
171 }
172
173 zero(p);
174 p.what = "cgroup";
175 p.where = where;
176 p.type = "cgroup";
177 p.options = controller;
178 p.flags = MS_NOSUID|MS_NOEXEC|MS_NODEV;
179 p.fatal = false;
180
181 r = mount_one(&p);
182 free(controller);
183 free(where);
184
185 if (r < 0)
186 goto finish;
187 }
188
189 r = 0;
190
191 finish:
192 fclose(f);
193
194 return r;
195 }
196
197 static int symlink_and_label(const char *old_path, const char *new_path) {
198 int r;
199
200 assert(old_path);
201 assert(new_path);
202
203 if ((r = label_symlinkfile_set(new_path)) < 0)
204 return r;
205
206 if (symlink(old_path, new_path) < 0)
207 r = -errno;
208
209 label_file_clear();
210
211 return r;
212 }
213
214 static int nftw_cb(
215 const char *fpath,
216 const struct stat *sb,
217 int tflag,
218 struct FTW *ftwbuf) {
219
220 /* No need to label /dev twice in a row... */
221 if (ftwbuf->level == 0)
222 return 0;
223
224 label_fix(fpath, true);
225 return 0;
226 };
227
228 int mount_setup(void) {
229
230 const char symlinks[] =
231 "/proc/kcore\0" "/dev/core\0"
232 "/proc/self/fd\0" "/dev/fd\0"
233 "/proc/self/fd/0\0" "/dev/stdin\0"
234 "/proc/self/fd/1\0" "/dev/stdout\0"
235 "/proc/self/fd/2\0" "/dev/stderr\0";
236
237 int r;
238 unsigned i;
239 const char *j, *k;
240
241 for (i = 0; i < ELEMENTSOF(mount_table); i ++)
242 if ((r = mount_one(mount_table+i)) < 0)
243 return r;
244
245 /* Nodes in devtmpfs and /run need to be manually updated for
246 * the appropriate labels, after mounting. The other virtual
247 * API file systems like /sys and /proc do not need that, they
248 * use the same label for all their files. */
249 if (unlink("/dev/.systemd-relabel-run-dev") >= 0) {
250 nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS);
251 nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS);
252 }
253
254 /* Create a few default symlinks, which are normally created
255 * by udevd, but some scripts might need them before we start
256 * udevd. */
257 NULSTR_FOREACH_PAIR(j, k, symlinks)
258 symlink_and_label(j, k);
259
260 /* Create a few directories we always want around */
261 mkdir("/run/systemd", 0755);
262
263 return mount_cgroup_controllers();
264 }