1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
|
/* SPDX-License-Identifier: LGPL-2.1-or-later */
#include <fcntl.h>
#include <sys/ioctl.h>
#include <sys/mount.h>
#include "errno-util.h"
#include "fd-util.h"
#include "fileio.h"
#include "missing_fs.h"
#include "missing_magic.h"
#include "missing_sched.h"
#include "namespace-util.h"
#include "process-util.h"
#include "stat-util.h"
#include "stdio-util.h"
#include "user-util.h"
const struct namespace_info namespace_info[] = {
[NAMESPACE_CGROUP] = { "cgroup", "ns/cgroup", CLONE_NEWCGROUP, },
[NAMESPACE_IPC] = { "ipc", "ns/ipc", CLONE_NEWIPC, },
[NAMESPACE_NET] = { "net", "ns/net", CLONE_NEWNET, },
/* So, the mount namespace flag is called CLONE_NEWNS for historical
* reasons. Let's expose it here under a more explanatory name: "mnt".
* This is in-line with how the kernel exposes namespaces in /proc/$PID/ns. */
[NAMESPACE_MOUNT] = { "mnt", "ns/mnt", CLONE_NEWNS, },
[NAMESPACE_PID] = { "pid", "ns/pid", CLONE_NEWPID, },
[NAMESPACE_USER] = { "user", "ns/user", CLONE_NEWUSER, },
[NAMESPACE_UTS] = { "uts", "ns/uts", CLONE_NEWUTS, },
[NAMESPACE_TIME] = { "time", "ns/time", CLONE_NEWTIME, },
{ /* Allow callers to iterate over the array without using _NAMESPACE_TYPE_MAX. */ },
};
#define pid_namespace_path(pid, type) procfs_file_alloca(pid, namespace_info[type].proc_path)
int namespace_open(pid_t pid, int *pidns_fd, int *mntns_fd, int *netns_fd, int *userns_fd, int *root_fd) {
_cleanup_close_ int pidnsfd = -1, mntnsfd = -1, netnsfd = -1, usernsfd = -1;
int rfd = -1;
assert(pid >= 0);
if (mntns_fd) {
const char *mntns;
mntns = pid_namespace_path(pid, NAMESPACE_MOUNT);
mntnsfd = open(mntns, O_RDONLY|O_NOCTTY|O_CLOEXEC);
if (mntnsfd < 0)
return -errno;
}
if (pidns_fd) {
const char *pidns;
pidns = pid_namespace_path(pid, NAMESPACE_PID);
pidnsfd = open(pidns, O_RDONLY|O_NOCTTY|O_CLOEXEC);
if (pidnsfd < 0)
return -errno;
}
if (netns_fd) {
const char *netns;
netns = pid_namespace_path(pid, NAMESPACE_NET);
netnsfd = open(netns, O_RDONLY|O_NOCTTY|O_CLOEXEC);
if (netnsfd < 0)
return -errno;
}
if (userns_fd) {
const char *userns;
userns = pid_namespace_path(pid, NAMESPACE_USER);
usernsfd = open(userns, O_RDONLY|O_NOCTTY|O_CLOEXEC);
if (usernsfd < 0 && errno != ENOENT)
return -errno;
}
if (root_fd) {
const char *root;
root = procfs_file_alloca(pid, "root");
rfd = open(root, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY);
if (rfd < 0)
return -errno;
}
if (pidns_fd)
*pidns_fd = TAKE_FD(pidnsfd);
if (mntns_fd)
*mntns_fd = TAKE_FD(mntnsfd);
if (netns_fd)
*netns_fd = TAKE_FD(netnsfd);
if (userns_fd)
*userns_fd = TAKE_FD(usernsfd);
if (root_fd)
*root_fd = TAKE_FD(rfd);
return 0;
}
int namespace_enter(int pidns_fd, int mntns_fd, int netns_fd, int userns_fd, int root_fd) {
int r;
if (userns_fd >= 0) {
/* Can't setns to your own userns, since then you could escalate from non-root to root in
* your own namespace, so check if namespaces are equal before attempting to enter. */
r = files_same(FORMAT_PROC_FD_PATH(userns_fd), "/proc/self/ns/user", 0);
if (r < 0)
return r;
if (r)
userns_fd = -1;
}
if (pidns_fd >= 0)
if (setns(pidns_fd, CLONE_NEWPID) < 0)
return -errno;
if (mntns_fd >= 0)
if (setns(mntns_fd, CLONE_NEWNS) < 0)
return -errno;
if (netns_fd >= 0)
if (setns(netns_fd, CLONE_NEWNET) < 0)
return -errno;
if (userns_fd >= 0)
if (setns(userns_fd, CLONE_NEWUSER) < 0)
return -errno;
if (root_fd >= 0) {
if (fchdir(root_fd) < 0)
return -errno;
if (chroot(".") < 0)
return -errno;
}
return reset_uid_gid();
}
int fd_is_ns(int fd, unsigned long nsflag) {
struct statfs s;
int r;
/* Checks whether the specified file descriptor refers to a namespace created by specifying nsflag in clone().
* On old kernels there's no nice way to detect that, hence on those we'll return a recognizable error (EUCLEAN),
* so that callers can handle this somewhat nicely.
*
* This function returns > 0 if the fd definitely refers to a network namespace, 0 if it definitely does not
* refer to a network namespace, -EUCLEAN if we can't determine, and other negative error codes on error. */
if (fstatfs(fd, &s) < 0)
return -errno;
if (!is_fs_type(&s, NSFS_MAGIC)) {
/* On really old kernels, there was no "nsfs", and network namespace sockets belonged to procfs
* instead. Handle that in a somewhat smart way. */
if (is_fs_type(&s, PROC_SUPER_MAGIC)) {
struct statfs t;
/* OK, so it is procfs. Let's see if our own network namespace is procfs, too. If so, then the
* passed fd might refer to a network namespace, but we can't know for sure. In that case,
* return a recognizable error. */
if (statfs("/proc/self/ns/net", &t) < 0)
return -errno;
if (s.f_type == t.f_type)
return -EUCLEAN; /* It's possible, we simply don't know */
}
return 0; /* No! */
}
r = ioctl(fd, NS_GET_NSTYPE);
if (r < 0) {
if (errno == ENOTTY) /* Old kernels didn't know this ioctl, let's also return a recognizable error in that case */
return -EUCLEAN;
return -errno;
}
return (unsigned long) r == nsflag;
}
int detach_mount_namespace(void) {
/* Detaches the mount namespace, disabling propagation from our namespace to the host */
if (unshare(CLONE_NEWNS) < 0)
return -errno;
return RET_NERRNO(mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL));
}
int userns_acquire(const char *uid_map, const char *gid_map) {
char path[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(pid_t) + 1];
_cleanup_(sigkill_waitp) pid_t pid = 0;
_cleanup_close_ int userns_fd = -1;
int r;
assert(uid_map);
assert(gid_map);
/* Forks off a process in a new userns, configures the specified uidmap/gidmap, acquires an fd to it,
* and then kills the process again. This way we have a userns fd that is not bound to any
* process. We can use that for file system mounts and similar. */
r = safe_fork("(sd-mkuserns)", FORK_CLOSE_ALL_FDS|FORK_DEATHSIG|FORK_NEW_USERNS, &pid);
if (r < 0)
return r;
if (r == 0)
/* Child. We do nothing here, just freeze until somebody kills us. */
freeze();
xsprintf(path, "/proc/" PID_FMT "/uid_map", pid);
r = write_string_file(path, uid_map, WRITE_STRING_FILE_DISABLE_BUFFER);
if (r < 0)
return log_error_errno(r, "Failed to write UID map: %m");
xsprintf(path, "/proc/" PID_FMT "/gid_map", pid);
r = write_string_file(path, gid_map, WRITE_STRING_FILE_DISABLE_BUFFER);
if (r < 0)
return log_error_errno(r, "Failed to write GID map: %m");
r = namespace_open(pid, NULL, NULL, NULL, &userns_fd, NULL);
if (r < 0)
return log_error_errno(r, "Failed to open userns fd: %m");
return TAKE_FD(userns_fd);
}
int in_same_namespace(pid_t pid1, pid_t pid2, NamespaceType type) {
const char *ns_path;
struct stat ns_st1, ns_st2;
if (pid1 == 0)
pid1 = getpid_cached();
if (pid2 == 0)
pid2 = getpid_cached();
if (pid1 == pid2)
return 1;
ns_path = pid_namespace_path(pid1, type);
if (stat(ns_path, &ns_st1) < 0)
return -errno;
ns_path = pid_namespace_path(pid2, type);
if (stat(ns_path, &ns_st2) < 0)
return -errno;
return stat_inode_same(&ns_st1, &ns_st2);
}
|
