1USER_NAMESPACES(7) Linux Programmer's Manual USER_NAMESPACES(7)
2
6 user_namespaces - overview of Linux user namespaces
7
9 For an overview of namespaces, see namespaces(7).
10 User namespaces isolate security-related identifiers and attributes, in
12 particular, user IDs and group IDs (see credentials(7)), the root di‐
13 rectory, keys (see keyrings(7)), and capabilities (see capabili‐
14 ties(7)). A process's user and group IDs can be different inside and
15 outside a user namespace. In particular, a process can have a normal
16 unprivileged user ID outside a user namespace while at the same time
17 having a user ID of 0 inside the namespace; in other words, the process
18 has full privileges for operations inside the user namespace, but is
19 unprivileged for operations outside the namespace.
20 Nested namespaces, namespace membership
22 User namespaces can be nested; that is, each user namespace—except the
23 initial ("root") namespace—has a parent user namespace, and can have
24 zero or more child user namespaces. The parent user namespace is the
25 user namespace of the process that creates the user namespace via a
26 call to unshare(2) or clone(2) with the CLONE_NEWUSER flag.
27 The kernel imposes (since version 3.11) a limit of 32 nested levels of
29 user namespaces. Calls to unshare(2) or clone(2) that would cause this
30 limit to be exceeded fail with the error EUSERS.
31 Each process is a member of exactly one user namespace. A process cre‐
33 ated via fork(2) or clone(2) without the CLONE_NEWUSER flag is a member
34 of the same user namespace as its parent. A single-threaded process
35 can join another user namespace with setns(2) if it has the CAP_SYS_AD‐
36 MIN in that namespace; upon doing so, it gains a full set of capabili‐
37 ties in that namespace.
38 A call to clone(2) or unshare(2) with the CLONE_NEWUSER flag makes the
40 new child process (for clone(2)) or the caller (for unshare(2)) a mem‐
41 ber of the new user namespace created by the call.
42 The NS_GET_PARENT ioctl(2) operation can be used to discover the
44 parental relationship between user namespaces; see ioctl_ns(2).
45 Capabilities
47 The child process created by clone(2) with the CLONE_NEWUSER flag
48 starts out with a complete set of capabilities in the new user name‐
49 space. Likewise, a process that creates a new user namespace using un‐
50 share(2) or joins an existing user namespace using setns(2) gains a
51 full set of capabilities in that namespace. On the other hand, that
52 process has no capabilities in the parent (in the case of clone(2)) or
53 previous (in the case of unshare(2) and setns(2)) user namespace, even
54 if the new namespace is created or joined by the root user (i.e., a
55 process with user ID 0 in the root namespace).
56 Note that a call to execve(2) will cause a process's capabilities to be
58 recalculated in the usual way (see capabilities(7)). Consequently, un‐
59 less the process has a user ID of 0 within the namespace, or the exe‐
60 cutable file has a nonempty inheritable capabilities mask, the process
61 will lose all capabilities. See the discussion of user and group ID
62 mappings, below.
63 A call to clone(2) or unshare(2) using the CLONE_NEWUSER flag or a call
65 to setns(2) that moves the caller into another user namespace sets the
66 "securebits" flags (see capabilities(7)) to their default values (all
67 flags disabled) in the child (for clone(2)) or caller (for unshare(2)
68 or setns(2)). Note that because the caller no longer has capabilities
69 in its original user namespace after a call to setns(2), it is not pos‐
70 sible for a process to reset its "securebits" flags while retaining its
71 user namespace membership by using a pair of setns(2) calls to move to
72 another user namespace and then return to its original user namespace.
73 The rules for determining whether or not a process has a capability in
75 a particular user namespace are as follows:
76 1. A process has a capability inside a user namespace if it is a member
78 of that namespace and it has the capability in its effective capa‐
79 bility set. A process can gain capabilities in its effective capa‐
80 bility set in various ways. For example, it may execute a set-user-
81 ID program or an executable with associated file capabilities. In
82 addition, a process may gain capabilities via the effect of
83 clone(2), unshare(2), or setns(2), as already described.
84 2. If a process has a capability in a user namespace, then it has that
86 capability in all child (and further removed descendant) namespaces
87 as well.
88 3. When a user namespace is created, the kernel records the effective
90 user ID of the creating process as being the "owner" of the name‐
91 space. A process that resides in the parent of the user namespace
92 and whose effective user ID matches the owner of the namespace has
93 all capabilities in the namespace. By virtue of the previous rule,
94 this means that the process has all capabilities in all further re‐
95 moved descendant user namespaces as well. The NS_GET_OWNER_UID
96 ioctl(2) operation can be used to discover the user ID of the owner
97 of the namespace; see ioctl_ns(2).
98 Effect of capabilities within a user namespace
100 Having a capability inside a user namespace permits a process to per‐
101 form operations (that require privilege) only on resources governed by
102 that namespace. In other words, having a capability in a user name‐
103 space permits a process to perform privileged operations on resources
104 that are governed by (nonuser) namespaces owned by (associated with)
105 the user namespace (see the next subsection).
106 On the other hand, there are many privileged operations that affect re‐
108 sources that are not associated with any namespace type, for example,
109 changing the system (i.e., calendar) time (governed by CAP_SYS_TIME),
110 loading a kernel module (governed by CAP_SYS_MODULE), and creating a
111 device (governed by CAP_MKNOD). Only a process with privileges in the
112 initial user namespace can perform such operations.
113 Holding CAP_SYS_ADMIN within the user namespace that owns a process's
115 mount namespace allows that process to create bind mounts and mount the
116 following types of filesystems:
117 * /proc (since Linux 3.8)
119 * /sys (since Linux 3.8)
120 * devpts (since Linux 3.9)
121 * tmpfs(5) (since Linux 3.9)
122 * ramfs (since Linux 3.9)
123 * mqueue (since Linux 3.9)
124 * bpf (since Linux 4.4)
125 * overlayfs (since Linux 5.11)
126 Holding CAP_SYS_ADMIN within the user namespace that owns a process's
128 cgroup namespace allows (since Linux 4.6) that process to the mount the
129 cgroup version 2 filesystem and cgroup version 1 named hierarchies
130 (i.e., cgroup filesystems mounted with the "none,name=" option).
131 Holding CAP_SYS_ADMIN within the user namespace that owns a process's
133 PID namespace allows (since Linux 3.8) that process to mount /proc
134 filesystems.
135 Note, however, that mounting block-based filesystems can be done only
137 by a process that holds CAP_SYS_ADMIN in the initial user namespace.
138 Interaction of user namespaces and other types of namespaces
140 Starting in Linux 3.8, unprivileged processes can create user name‐
141 spaces, and the other types of namespaces can be created with just the
142 CAP_SYS_ADMIN capability in the caller's user namespace.
143 When a nonuser namespace is created, it is owned by the user namespace
145 in which the creating process was a member at the time of the creation
146 of the namespace. Privileged operations on resources governed by the
147 nonuser namespace require that the process has the necessary capabili‐
148 ties in the user namespace that owns the nonuser namespace.
149 If CLONE_NEWUSER is specified along with other CLONE_NEW* flags in a
151 single clone(2) or unshare(2) call, the user namespace is guaranteed to
152 be created first, giving the child (clone(2)) or caller (unshare(2))
153 privileges over the remaining namespaces created by the call. Thus, it
154 is possible for an unprivileged caller to specify this combination of
155 flags.
156 When a new namespace (other than a user namespace) is created via
158 clone(2) or unshare(2), the kernel records the user namespace of the
159 creating process as the owner of the new namespace. (This association
160 can't be changed.) When a process in the new namespace subsequently
161 performs privileged operations that operate on global resources iso‐
162 lated by the namespace, the permission checks are performed according
163 to the process's capabilities in the user namespace that the kernel as‐
164 sociated with the new namespace. For example, suppose that a process
165 attempts to change the hostname (sethostname(2)), a resource governed
166 by the UTS namespace. In this case, the kernel will determine which
167 user namespace owns the process's UTS namespace, and check whether the
168 process has the required capability (CAP_SYS_ADMIN) in that user name‐
169 space.
170 The NS_GET_USERNS ioctl(2) operation can be used to discover the user
172 namespace that owns a nonuser namespace; see ioctl_ns(2).
173 User and group ID mappings: uid_map and gid_map
175 When a user namespace is created, it starts out without a mapping of
176 user IDs (group IDs) to the parent user namespace. The
177 /proc/[pid]/uid_map and /proc/[pid]/gid_map files (available since
178 Linux 3.5) expose the mappings for user and group IDs inside the user
179 namespace for the process pid. These files can be read to view the
180 mappings in a user namespace and written to (once) to define the map‐
181 pings.
182 The description in the following paragraphs explains the details for
184 uid_map; gid_map is exactly the same, but each instance of "user ID" is
185 replaced by "group ID".
186 The uid_map file exposes the mapping of user IDs from the user name‐
188 space of the process pid to the user namespace of the process that
189 opened uid_map (but see a qualification to this point below). In other
190 words, processes that are in different user namespaces will potentially
191 see different values when reading from a particular uid_map file, de‐
192 pending on the user ID mappings for the user namespaces of the reading
193 processes.
194 Each line in the uid_map file specifies a 1-to-1 mapping of a range of
196 contiguous user IDs between two user namespaces. (When a user name‐
197 space is first created, this file is empty.) The specification in each
198 line takes the form of three numbers delimited by white space. The
199 first two numbers specify the starting user ID in each of the two user
200 namespaces. The third number specifies the length of the mapped range.
201 In detail, the fields are interpreted as follows:
202 (1) The start of the range of user IDs in the user namespace of the
204 process pid.
205 (2) The start of the range of user IDs to which the user IDs specified
207 by field one map. How field two is interpreted depends on whether
208 the process that opened uid_map and the process pid are in the same
209 user namespace, as follows:
210 a) If the two processes are in different user namespaces: field two
212 is the start of a range of user IDs in the user namespace of the
213 process that opened uid_map.
214 b) If the two processes are in the same user namespace: field two
216 is the start of the range of user IDs in the parent user name‐
217 space of the process pid. This case enables the opener of
218 uid_map (the common case here is opening /proc/self/uid_map) to
219 see the mapping of user IDs into the user namespace of the
220 process that created this user namespace.
221 (3) The length of the range of user IDs that is mapped between the two
223 user namespaces.
224 System calls that return user IDs (group IDs)—for example, getuid(2),
226 getgid(2), and the credential fields in the structure returned by
227 stat(2)—return the user ID (group ID) mapped into the caller's user
228 namespace.
229 When a process accesses a file, its user and group IDs are mapped into
231 the initial user namespace for the purpose of permission checking and
232 assigning IDs when creating a file. When a process retrieves file user
233 and group IDs via stat(2), the IDs are mapped in the opposite direc‐
234 tion, to produce values relative to the process user and group ID map‐
235 pings.
236 The initial user namespace has no parent namespace, but, for consis‐
238 tency, the kernel provides dummy user and group ID mapping files for
239 this namespace. Looking at the uid_map file (gid_map is the same) from
240 a shell in the initial namespace shows:
241 $ cat /proc/$$/uid_map
243 0 0 4294967295
244 This mapping tells us that the range starting at user ID 0 in this
246 namespace maps to a range starting at 0 in the (nonexistent) parent
247 namespace, and the length of the range is the largest 32-bit unsigned
248 integer. This leaves 4294967295 (the 32-bit signed -1 value) unmapped.
249 This is deliberate: (uid_t) -1 is used in several interfaces (e.g., se‐
250 treuid(2)) as a way to specify "no user ID". Leaving (uid_t) -1 un‐
251 mapped and unusable guarantees that there will be no confusion when us‐
252 ing these interfaces.
253 Defining user and group ID mappings: writing to uid_map and gid_map
255 After the creation of a new user namespace, the uid_map file of one of
256 the processes in the namespace may be written to once to define the
257 mapping of user IDs in the new user namespace. An attempt to write
258 more than once to a uid_map file in a user namespace fails with the er‐
259 ror EPERM. Similar rules apply for gid_map files.
260 The lines written to uid_map (gid_map) must conform to the following
262 validity rules:
263 * The three fields must be valid numbers, and the last field must be
265 greater than 0.
266 * Lines are terminated by newline characters.
268 * There is a limit on the number of lines in the file. In Linux 4.14
270 and earlier, this limit was (arbitrarily) set at 5 lines. Since
271 Linux 4.15, the limit is 340 lines. In addition, the number of
272 bytes written to the file must be less than the system page size,
273 and the write must be performed at the start of the file (i.e.,
274 lseek(2) and pwrite(2) can't be used to write to nonzero offsets in
275 the file).
276 * The range of user IDs (group IDs) specified in each line cannot
278 overlap with the ranges in any other lines. In the initial imple‐
279 mentation (Linux 3.8), this requirement was satisfied by a simplis‐
280 tic implementation that imposed the further requirement that the
281 values in both field 1 and field 2 of successive lines must be in
282 ascending numerical order, which prevented some otherwise valid maps
283 from being created. Linux 3.9 and later fix this limitation, allow‐
284 ing any valid set of nonoverlapping maps.
285 * At least one line must be written to the file.
287 Writes that violate the above rules fail with the error EINVAL.
289 In order for a process to write to the /proc/[pid]/uid_map
291 (/proc/[pid]/gid_map) file, all of the following permission require‐
292 ments must be met:
293 1. The writing process must have the CAP_SETUID (CAP_SETGID) capability
295 in the user namespace of the process pid.
296 2. The writing process must either be in the user namespace of the
298 process pid or be in the parent user namespace of the process pid.
299 3. The mapped user IDs (group IDs) must in turn have a mapping in the
301 parent user namespace.
302 4. If updating /proc/[pid]/uid_map to create a mapping that maps UID 0
304 in the parent namespace, then one of the following must be true:
305 * if writing process is in the parent user namespace, then it must
307 have the CAP_SETFCAP capability in that user namespace; or
308 * if the writing process is in the child user namespace, then the
310 process that created the user namespace must have had the
311 CAP_SETFCAP capability when the namespace was created.
312 This rule has been in place since Linux 5.12. It eliminates an ear‐
314 lier security bug whereby a UID 0 process that lacks the CAP_SETFCAP
315 capability, which is needed to create a binary with namespaced file
316 capabilities (as described in capabilities(7)), could nevertheless
317 create such a binary, by the following steps:
318 * Create a new user namespace with the identity mapping (i.e., UID
320 0 in the new user namespace maps to UID 0 in the parent name‐
321 space), so that UID 0 in both namespaces is equivalent to the
322 same root user ID.
323 * Since the child process has the CAP_SETFCAP capability, it could
325 create a binary with namespaced file capabilities that would then
326 be effective in the parent user namespace (because the root user
327 IDs are the same in the two namespaces).
328 5. One of the following two cases applies:
330 * Either the writing process has the CAP_SETUID (CAP_SETGID) capa‐
332 bility in the parent user namespace.
333 + No further restrictions apply: the process can make mappings
335 to arbitrary user IDs (group IDs) in the parent user name‐
336 space.
337 * Or otherwise all of the following restrictions apply:
339 + The data written to uid_map (gid_map) must consist of a single
341 line that maps the writing process's effective user ID (group
342 ID) in the parent user namespace to a user ID (group ID) in
343 the user namespace.
344 + The writing process must have the same effective user ID as
346 the process that created the user namespace.
347 + In the case of gid_map, use of the setgroups(2) system call
349 must first be denied by writing "deny" to the /proc/[pid]/set‐
350 groups file (see below) before writing to gid_map.
351 Writes that violate the above rules fail with the error EPERM.
353 Project ID mappings: projid_map
355 Similarly to user and group ID mappings, it is possible to create
356 project ID mappings for a user namespace. (Project IDs are used for
357 disk quotas; see setquota(8) and quotactl(2).)
358 Project ID mappings are defined by writing to the /proc/[pid]/pro‐
360 jid_map file (present since Linux 3.7).
361 The validity rules for writing to the /proc/[pid]/projid_map file are
363 as for writing to the uid_map file; violation of these rules causes
364 write(2) to fail with the error EINVAL.
365 The permission rules for writing to the /proc/[pid]/projid_map file are
367 as follows:
368 1. The writing process must either be in the user namespace of the
370 process pid or be in the parent user namespace of the process pid.
371 2. The mapped project IDs must in turn have a mapping in the parent
373 user namespace.
374 Violation of these rules causes write(2) to fail with the error EPERM.
376 Interaction with system calls that change process UIDs or GIDs
378 In a user namespace where the uid_map file has not been written, the
379 system calls that change user IDs will fail. Similarly, if the gid_map
380 file has not been written, the system calls that change group IDs will
381 fail. After the uid_map and gid_map files have been written, only the
382 mapped values may be used in system calls that change user and group
383 IDs.
384 For user IDs, the relevant system calls include setuid(2), setfsuid(2),
386 setreuid(2), and setresuid(2). For group IDs, the relevant system
387 calls include setgid(2), setfsgid(2), setregid(2), setresgid(2), and
388 setgroups(2).
389 Writing "deny" to the /proc/[pid]/setgroups file before writing to
391 /proc/[pid]/gid_map will permanently disable setgroups(2) in a user
392 namespace and allow writing to /proc/[pid]/gid_map without having the
393 CAP_SETGID capability in the parent user namespace.
394 The /proc/[pid]/setgroups file
396 The /proc/[pid]/setgroups file displays the string "allow" if processes
397 in the user namespace that contains the process pid are permitted to
398 employ the setgroups(2) system call; it displays "deny" if setgroups(2)
399 is not permitted in that user namespace. Note that regardless of the
400 value in the /proc/[pid]/setgroups file (and regardless of the
401 process's capabilities), calls to setgroups(2) are also not permitted
402 if /proc/[pid]/gid_map has not yet been set.
403 A privileged process (one with the CAP_SYS_ADMIN capability in the
405 namespace) may write either of the strings "allow" or "deny" to this
406 file before writing a group ID mapping for this user namespace to the
407 file /proc/[pid]/gid_map. Writing the string "deny" prevents any
408 process in the user namespace from employing setgroups(2).
409 The essence of the restrictions described in the preceding paragraph is
411 that it is permitted to write to /proc/[pid]/setgroups only so long as
412 calling setgroups(2) is disallowed because /proc/[pid]/gid_map has not
413 been set. This ensures that a process cannot transition from a state
414 where setgroups(2) is allowed to a state where setgroups(2) is denied;
415 a process can transition only from setgroups(2) being disallowed to
416 setgroups(2) being allowed.
417 The default value of this file in the initial user namespace is "al‐
419 low".
420 Once /proc/[pid]/gid_map has been written to (which has the effect of
422 enabling setgroups(2) in the user namespace), it is no longer possible
423 to disallow setgroups(2) by writing "deny" to /proc/[pid]/setgroups
424 (the write fails with the error EPERM).
425 A child user namespace inherits the /proc/[pid]/setgroups setting from
427 its parent.
428 If the setgroups file has the value "deny", then the setgroups(2) sys‐
430 tem call can't subsequently be reenabled (by writing "allow" to the
431 file) in this user namespace. (Attempts to do so fail with the error
432 EPERM.) This restriction also propagates down to all child user name‐
433 spaces of this user namespace.
434 The /proc/[pid]/setgroups file was added in Linux 3.19, but was back‐
436 ported to many earlier stable kernel series, because it addresses a se‐
437 curity issue. The issue concerned files with permissions such as
438 "rwx---rwx". Such files give fewer permissions to "group" than they do
439 to "other". This means that dropping groups using setgroups(2) might
440 allow a process file access that it did not formerly have. Before the
441 existence of user namespaces this was not a concern, since only a priv‐
442 ileged process (one with the CAP_SETGID capability) could call set‐
443 groups(2). However, with the introduction of user namespaces, it be‐
444 came possible for an unprivileged process to create a new namespace in
445 which the user had all privileges. This then allowed formerly unprivi‐
446 leged users to drop groups and thus gain file access that they did not
447 previously have. The /proc/[pid]/setgroups file was added to address
448 this security issue, by denying any pathway for an unprivileged process
449 to drop groups with setgroups(2).
450 Unmapped user and group IDs
452 There are various places where an unmapped user ID (group ID) may be
453 exposed to user space. For example, the first process in a new user
454 namespace may call getuid(2) before a user ID mapping has been defined
455 for the namespace. In most such cases, an unmapped user ID is con‐
456 verted to the overflow user ID (group ID); the default value for the
457 overflow user ID (group ID) is 65534. See the descriptions of
458 /proc/sys/kernel/overflowuid and /proc/sys/kernel/overflowgid in
459 proc(5).
460 The cases where unmapped IDs are mapped in this fashion include system
462 calls that return user IDs (getuid(2), getgid(2), and similar), creden‐
463 tials passed over a UNIX domain socket, credentials returned by
464 stat(2), waitid(2), and the System V IPC "ctl" IPC_STAT operations,
465 credentials exposed by /proc/[pid]/status and the files in
466 /proc/sysvipc/*, credentials returned via the si_uid field in the sig‐
467 info_t received with a signal (see sigaction(2)), credentials written
468 to the process accounting file (see acct(5)), and credentials returned
469 with POSIX message queue notifications (see mq_notify(3)).
470 There is one notable case where unmapped user and group IDs are not
472 converted to the corresponding overflow ID value. When viewing a
473 uid_map or gid_map file in which there is no mapping for the second
474 field, that field is displayed as 4294967295 (-1 as an unsigned inte‐
475 ger).
476 Accessing files
478 In order to determine permissions when an unprivileged process accesses
479 a file, the process credentials (UID, GID) and the file credentials are
480 in effect mapped back to what they would be in the initial user name‐
481 space and then compared to determine the permissions that the process
482 has on the file. The same is also of other objects that employ the
483 credentials plus permissions mask accessibility model, such as System V
484 IPC objects
485 Operation of file-related capabilities
487 Certain capabilities allow a process to bypass various kernel-enforced
488 restrictions when performing operations on files owned by other users
489 or groups. These capabilities are: CAP_CHOWN, CAP_DAC_OVERRIDE,
490 CAP_DAC_READ_SEARCH, CAP_FOWNER, and CAP_FSETID.
491 Within a user namespace, these capabilities allow a process to bypass
493 the rules if the process has the relevant capability over the file,
494 meaning that:
495 * the process has the relevant effective capability in its user name‐
497 space; and
498 * the file's user ID and group ID both have valid mappings in the user
500 namespace.
501 The CAP_FOWNER capability is treated somewhat exceptionally: it allows
503 a process to bypass the corresponding rules so long as at least the
504 file's user ID has a mapping in the user namespace (i.e., the file's
505 group ID does not need to have a valid mapping).
506 Set-user-ID and set-group-ID programs
508 When a process inside a user namespace executes a set-user-ID (set-
509 group-ID) program, the process's effective user (group) ID inside the
510 namespace is changed to whatever value is mapped for the user (group)
511 ID of the file. However, if either the user or the group ID of the
512 file has no mapping inside the namespace, the set-user-ID (set-group-
513 ID) bit is silently ignored: the new program is executed, but the
514 process's effective user (group) ID is left unchanged. (This mirrors
515 the semantics of executing a set-user-ID or set-group-ID program that
516 resides on a filesystem that was mounted with the MS_NOSUID flag, as
517 described in mount(2).)
518 Miscellaneous
520 When a process's user and group IDs are passed over a UNIX domain
521 socket to a process in a different user namespace (see the description
522 of SCM_CREDENTIALS in unix(7)), they are translated into the corre‐
523 sponding values as per the receiving process's user and group ID map‐
524 pings.
525
527 Namespaces are a Linux-specific feature.
528
530 Over the years, there have been a lot of features that have been added
531 to the Linux kernel that have been made available only to privileged
532 users because of their potential to confuse set-user-ID-root applica‐
533 tions. In general, it becomes safe to allow the root user in a user
534 namespace to use those features because it is impossible, while in a
535 user namespace, to gain more privilege than the root user of a user
536 namespace has.
537 Global root
539 The term "global root" is sometimes used as a shorthand for user ID 0
540 in the initial user namespace.
541 Availability
543 Use of user namespaces requires a kernel that is configured with the
544 CONFIG_USER_NS option. User namespaces require support in a range of
545 subsystems across the kernel. When an unsupported subsystem is config‐
546 ured into the kernel, it is not possible to configure user namespaces
547 support.
548 As at Linux 3.8, most relevant subsystems supported user namespaces,
550 but a number of filesystems did not have the infrastructure needed to
551 map user and group IDs between user namespaces. Linux 3.9 added the
552 required infrastructure support for many of the remaining unsupported
553 filesystems (Plan 9 (9P), Andrew File System (AFS), Ceph, CIFS, CODA,
554 NFS, and OCFS2). Linux 3.12 added support for the last of the unsup‐
555 ported major filesystems, XFS.
556
558 The program below is designed to allow experimenting with user name‐
559 spaces, as well as other types of namespaces. It creates namespaces as
560 specified by command-line options and then executes a command inside
561 those namespaces. The comments and usage() function inside the program
562 provide a full explanation of the program. The following shell session
563 demonstrates its use.
564 First, we look at the run-time environment:
566 $ uname -rs # Need Linux 3.8 or later
568 Linux 3.8.0
569 $ id -u # Running as unprivileged user
570 1000
571 $ id -g
572 1000
573 Now start a new shell in new user (-U), mount (-m), and PID (-p) name‐
575 spaces, with user ID (-M) and group ID (-G) 1000 mapped to 0 inside the
576 user namespace:
577 $ ./userns_child_exec -p -m -U -M '0 1000 1' -G '0 1000 1' bash
579 The shell has PID 1, because it is the first process in the new PID
581 namespace:
582 bash$ echo $$
584 1
585 Mounting a new /proc filesystem and listing all of the processes visi‐
587 ble in the new PID namespace shows that the shell can't see any pro‐
588 cesses outside the PID namespace:
589 bash$ mount -t proc proc /proc
591 bash$ ps ax
592 PID TTY STAT TIME COMMAND
593 1 pts/3 S 0:00 bash
594 22 pts/3 R+ 0:00 ps ax
595 Inside the user namespace, the shell has user and group ID 0, and a
597 full set of permitted and effective capabilities:
598 bash$ cat /proc/$$/status | egrep '^[UG]id'
600 Uid: 0 0 0 0
601 Gid: 0 0 0 0
602 bash$ cat /proc/$$/status | egrep '^Cap(Prm|Inh|Eff)'
603 CapInh: 0000000000000000
604 CapPrm: 0000001fffffffff
605 CapEff: 0000001fffffffff
606 Program source
608 /* userns_child_exec.c
610 Licensed under GNU General Public License v2 or later
612 Create a child process that executes a shell command in new
614 namespace(s); allow UID and GID mappings to be specified when
615 creating a user namespace.
616 */
617 #define _GNU_SOURCE
618 #include <sched.h>
619 #include <unistd.h>
620 #include <stdint.h>
621 #include <stdlib.h>
622 #include <sys/wait.h>
623 #include <signal.h>
624 #include <fcntl.h>
625 #include <stdio.h>
626 #include <string.h>
627 #include <limits.h>
628 #include <errno.h>
629 /* A simple error-handling function: print an error message based
631 on the value in 'errno' and terminate the calling process. */
632 #define errExit(msg) do { perror(msg); exit(EXIT_FAILURE); \
634 } while (0)
635 struct child_args {
637 char **argv; /* Command to be executed by child, with args */
638 int pipe_fd[2]; /* Pipe used to synchronize parent and child */
639 };
640 static int verbose;
642 static void
644 usage(char *pname)
645 {
646 fprintf(stderr, "Usage: %s [options] cmd [arg...]\n\n", pname);
647 fprintf(stderr, "Create a child process that executes a shell "
648 "command in a new user namespace,\n"
649 "and possibly also other new namespace(s).\n\n");
650 fprintf(stderr, "Options can be:\n\n");
651 #define fpe(str) fprintf(stderr, " %s", str);
652 fpe("-i New IPC namespace\n");
653 fpe("-m New mount namespace\n");
654 fpe("-n New network namespace\n");
655 fpe("-p New PID namespace\n");
656 fpe("-u New UTS namespace\n");
657 fpe("-U New user namespace\n");
658 fpe("-M uid_map Specify UID map for user namespace\n");
659 fpe("-G gid_map Specify GID map for user namespace\n");
660 fpe("-z Map user's UID and GID to 0 in user namespace\n");
661 fpe(" (equivalent to: -M '0 <uid> 1' -G '0 <gid> 1')\n");
662 fpe("-v Display verbose messages\n");
663 fpe("\n");
664 fpe("If -z, -M, or -G is specified, -U is required.\n");
665 fpe("It is not permitted to specify both -z and either -M or -G.\n");
666 fpe("\n");
667 fpe("Map strings for -M and -G consist of records of the form:\n");
668 fpe("\n");
669 fpe(" ID-inside-ns ID-outside-ns len\n");
670 fpe("\n");
671 fpe("A map string can contain multiple records, separated"
672 " by commas;\n");
673 fpe("the commas are replaced by newlines before writing"
674 " to map files.\n");
675 exit(EXIT_FAILURE);
677 }
678 /* Update the mapping file 'map_file', with the value provided in
680 'mapping', a string that defines a UID or GID mapping. A UID or
681 GID mapping consists of one or more newline-delimited records
682 of the form:
683 ID_inside-ns ID-outside-ns length
685 Requiring the user to supply a string that contains newlines is
687 of course inconvenient for command-line use. Thus, we permit the
688 use of commas to delimit records in this string, and replace them
689 with newlines before writing the string to the file. */
690 static void
692 update_map(char *mapping, char *map_file)
693 {
694 int fd;
695 size_t map_len; /* Length of 'mapping' */
696 /* Replace commas in mapping string with newlines. */
698 map_len = strlen(mapping);
700 for (int j = 0; j < map_len; j++)
701 if (mapping[j] == ',')
702 mapping[j] = '\n';
703 fd = open(map_file, O_RDWR);
705 if (fd == -1) {
706 fprintf(stderr, "ERROR: open %s: %s\n", map_file,
707 strerror(errno));
708 exit(EXIT_FAILURE);
709 }
710 if (write(fd, mapping, map_len) != map_len) {
712 fprintf(stderr, "ERROR: write %s: %s\n", map_file,
713 strerror(errno));
714 exit(EXIT_FAILURE);
715 }
716 close(fd);
718 }
719 /* Linux 3.19 made a change in the handling of setgroups(2) and the
721 'gid_map' file to address a security issue. The issue allowed
722 *unprivileged* users to employ user namespaces in order to drop
723 The upshot of the 3.19 changes is that in order to update the
724 'gid_maps' file, use of the setgroups() system call in this
725 user namespace must first be disabled by writing "deny" to one of
726 the /proc/PID/setgroups files for this namespace. That is the
727 purpose of the following function. */
728 static void
730 proc_setgroups_write(pid_t child_pid, char *str)
731 {
732 char setgroups_path[PATH_MAX];
733 int fd;
734 snprintf(setgroups_path, PATH_MAX, "/proc/%jd/setgroups",
736 (intmax_t) child_pid);
737 fd = open(setgroups_path, O_RDWR);
739 if (fd == -1) {
740 /* We may be on a system that doesn't support
742 /proc/PID/setgroups. In that case, the file won't exist,
743 and the system won't impose the restrictions that Linux 3.19
744 added. That's fine: we don't need to do anything in order
745 to permit 'gid_map' to be updated.
746 However, if the error from open() was something other than
748 the ENOENT error that is expected for that case, let the
749 user know. */
750 if (errno != ENOENT)
752 fprintf(stderr, "ERROR: open %s: %s\n", setgroups_path,
753 strerror(errno));
754 return;
755 }
756 if (write(fd, str, strlen(str)) == -1)
758 fprintf(stderr, "ERROR: write %s: %s\n", setgroups_path,
759 strerror(errno));
760 close(fd);
762 }
763 static int /* Start function for cloned child */
765 childFunc(void *arg)
766 {
767 struct child_args *args = arg;
768 char ch;
769 /* Wait until the parent has updated the UID and GID mappings.
771 See the comment in main(). We wait for end of file on a
772 pipe that will be closed by the parent process once it has
773 updated the mappings. */
774 close(args->pipe_fd[1]); /* Close our descriptor for the write
776 end of the pipe so that we see EOF
777 when parent closes its descriptor. */
778 if (read(args->pipe_fd[0], &ch, 1) != 0) {
779 fprintf(stderr,
780 "Failure in child: read from pipe returned != 0\n");
781 exit(EXIT_FAILURE);
782 }
783 close(args->pipe_fd[0]);
785 /* Execute a shell command. */
787 printf("About to exec %s\n", args->argv[0]);
789 execvp(args->argv[0], args->argv);
790 errExit("execvp");
791 }
792 #define STACK_SIZE (1024 * 1024)
794 static char child_stack[STACK_SIZE]; /* Space for child's stack */
796 int
798 main(int argc, char *argv[])
799 {
800 int flags, opt, map_zero;
801 pid_t child_pid;
802 struct child_args args;
803 char *uid_map, *gid_map;
804 const int MAP_BUF_SIZE = 100;
805 char map_buf[MAP_BUF_SIZE];
806 char map_path[PATH_MAX];
807 /* Parse command-line options. The initial '+' character in
809 the final getopt() argument prevents GNU-style permutation
810 of command-line options. That's useful, since sometimes
811 the 'command' to be executed by this program itself
812 has command-line options. We don't want getopt() to treat
813 those as options to this program. */
814 flags = 0;
816 verbose = 0;
817 gid_map = NULL;
818 uid_map = NULL;
819 map_zero = 0;
820 while ((opt = getopt(argc, argv, "+imnpuUM:G:zv")) != -1) {
821 switch (opt) {
822 case 'i': flags |= CLONE_NEWIPC; break;
823 case 'm': flags |= CLONE_NEWNS; break;
824 case 'n': flags |= CLONE_NEWNET; break;
825 case 'p': flags |= CLONE_NEWPID; break;
826 case 'u': flags |= CLONE_NEWUTS; break;
827 case 'v': verbose = 1; break;
828 case 'z': map_zero = 1; break;
829 case 'M': uid_map = optarg; break;
830 case 'G': gid_map = optarg; break;
831 case 'U': flags |= CLONE_NEWUSER; break;
832 default: usage(argv[0]);
833 }
834 }
835 /* -M or -G without -U is nonsensical */
837 if (((uid_map != NULL || gid_map != NULL || map_zero) &&
839 !(flags & CLONE_NEWUSER)) ||
840 (map_zero && (uid_map != NULL || gid_map != NULL)))
841 usage(argv[0]);
842 args.argv = &argv[optind];
844 /* We use a pipe to synchronize the parent and child, in order to
846 ensure that the parent sets the UID and GID maps before the child
847 calls execve(). This ensures that the child maintains its
848 capabilities during the execve() in the common case where we
849 want to map the child's effective user ID to 0 in the new user
850 namespace. Without this synchronization, the child would lose
851 its capabilities if it performed an execve() with nonzero
852 user IDs (see the capabilities(7) man page for details of the
853 transformation of a process's capabilities during execve()). */
854 if (pipe(args.pipe_fd) == -1)
856 errExit("pipe");
857 /* Create the child in new namespace(s). */
859 child_pid = clone(childFunc, child_stack + STACK_SIZE,
861 flags | SIGCHLD, &args);
862 if (child_pid == -1)
863 errExit("clone");
864 /* Parent falls through to here. */
866 if (verbose)
868 printf("%s: PID of child created by clone() is %jd\n",
869 argv[0], (intmax_t) child_pid);
870 /* Update the UID and GID maps in the child. */
872 if (uid_map != NULL || map_zero) {
874 snprintf(map_path, PATH_MAX, "/proc/%jd/uid_map",
875 (intmax_t) child_pid);
876 if (map_zero) {
877 snprintf(map_buf, MAP_BUF_SIZE, "0 %jd 1",
878 (intmax_t) getuid());
879 uid_map = map_buf;
880 }
881 update_map(uid_map, map_path);
882 }
883 if (gid_map != NULL || map_zero) {
885 proc_setgroups_write(child_pid, "deny");
886 snprintf(map_path, PATH_MAX, "/proc/%jd/gid_map",
888 (intmax_t) child_pid);
889 if (map_zero) {
890 snprintf(map_buf, MAP_BUF_SIZE, "0 %ld 1",
891 (intmax_t) getgid());
892 gid_map = map_buf;
893 }
894 update_map(gid_map, map_path);
895 }
896 /* Close the write end of the pipe, to signal to the child that we
898 have updated the UID and GID maps. */
899 close(args.pipe_fd[1]);
901 if (waitpid(child_pid, NULL, 0) == -1) /* Wait for child */
903 errExit("waitpid");
904 if (verbose)
906 printf("%s: terminating\n", argv[0]);
907 exit(EXIT_SUCCESS);
909 }
910
912 newgidmap(1), newuidmap(1), clone(2), ptrace(2), setns(2), unshare(2),
913 proc(5), subgid(5), subuid(5), capabilities(7), cgroup_namespaces(7),
914 credentials(7), namespaces(7), pid_namespaces(7)
915 The kernel source file Documentation/admin-guide/namespaces/re‐
917 source-control.rst.
918
920 This page is part of release 5.13 of the Linux man-pages project. A
921 description of the project, information about reporting bugs, and the
922 latest version of this page, can be found at
923 https://www.kernel.org/doc/man-pages/.
924Linux 2021-08-27 USER_NAMESPACES(7)