1user_namespaces(7) Miscellaneous Information 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 Linux 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 A task that changes one of its effective IDs will have its dumpability
47 reset to the value in /proc/sys/fs/suid_dumpable. This may affect the
48 ownership of proc files of child processes and may thus cause the par‐
49 ent to lack the permissions to write to mapping files of child pro‐
50 cesses running in a new user namespace. In such cases making the par‐
51 ent process dumpable, using PR_SET_DUMPABLE in a call to prctl(2), be‐
52 fore creating a child process in a new user namespace may rectify this
53 problem. See prctl(2) and proc(5) for details on how ownership is af‐
54 fected.
55 Capabilities
57 The child process created by clone(2) with the CLONE_NEWUSER flag
58 starts out with a complete set of capabilities in the new user name‐
59 space. Likewise, a process that creates a new user namespace using un‐
60 share(2) or joins an existing user namespace using setns(2) gains a
61 full set of capabilities in that namespace. On the other hand, that
62 process has no capabilities in the parent (in the case of clone(2)) or
63 previous (in the case of unshare(2) and setns(2)) user namespace, even
64 if the new namespace is created or joined by the root user (i.e., a
65 process with user ID 0 in the root namespace).
66 Note that a call to execve(2) will cause a process's capabilities to be
68 recalculated in the usual way (see capabilities(7)). Consequently, un‐
69 less the process has a user ID of 0 within the namespace, or the exe‐
70 cutable file has a nonempty inheritable capabilities mask, the process
71 will lose all capabilities. See the discussion of user and group ID
72 mappings, below.
73 A call to clone(2) or unshare(2) using the CLONE_NEWUSER flag or a call
75 to setns(2) that moves the caller into another user namespace sets the
76 "securebits" flags (see capabilities(7)) to their default values (all
77 flags disabled) in the child (for clone(2)) or caller (for unshare(2)
78 or setns(2)). Note that because the caller no longer has capabilities
79 in its original user namespace after a call to setns(2), it is not pos‐
80 sible for a process to reset its "securebits" flags while retaining its
81 user namespace membership by using a pair of setns(2) calls to move to
82 another user namespace and then return to its original user namespace.
83 The rules for determining whether or not a process has a capability in
85 a particular user namespace are as follows:
86 • A process has a capability inside a user namespace if it is a member
88 of that namespace and it has the capability in its effective capa‐
89 bility set. A process can gain capabilities in its effective capa‐
90 bility set in various ways. For example, it may execute a set-user-
91 ID program or an executable with associated file capabilities. In
92 addition, a process may gain capabilities via the effect of
93 clone(2), unshare(2), or setns(2), as already described.
94 • If a process has a capability in a user namespace, then it has that
96 capability in all child (and further removed descendant) namespaces
97 as well.
98 • When a user namespace is created, the kernel records the effective
100 user ID of the creating process as being the "owner" of the name‐
101 space. A process that resides in the parent of the user namespace
102 and whose effective user ID matches the owner of the namespace has
103 all capabilities in the namespace. By virtue of the previous rule,
104 this means that the process has all capabilities in all further re‐
105 moved descendant user namespaces as well. The NS_GET_OWNER_UID
106 ioctl(2) operation can be used to discover the user ID of the owner
107 of the namespace; see ioctl_ns(2).
108 Effect of capabilities within a user namespace
110 Having a capability inside a user namespace permits a process to per‐
111 form operations (that require privilege) only on resources governed by
112 that namespace. In other words, having a capability in a user name‐
113 space permits a process to perform privileged operations on resources
114 that are governed by (nonuser) namespaces owned by (associated with)
115 the user namespace (see the next subsection).
116 On the other hand, there are many privileged operations that affect re‐
118 sources that are not associated with any namespace type, for example,
119 changing the system (i.e., calendar) time (governed by CAP_SYS_TIME),
120 loading a kernel module (governed by CAP_SYS_MODULE), and creating a
121 device (governed by CAP_MKNOD). Only a process with privileges in the
122 initial user namespace can perform such operations.
123 Holding CAP_SYS_ADMIN within the user namespace that owns a process's
125 mount namespace allows that process to create bind mounts and mount the
126 following types of filesystems:
127 • /proc (since Linux 3.8)
129 • /sys (since Linux 3.8)
130 • devpts (since Linux 3.9)
131 • tmpfs(5) (since Linux 3.9)
132 • ramfs (since Linux 3.9)
133 • mqueue (since Linux 3.9)
134 • bpf (since Linux 4.4)
135 • overlayfs (since Linux 5.11)
136 Holding CAP_SYS_ADMIN within the user namespace that owns a process's
138 cgroup namespace allows (since Linux 4.6) that process to the mount the
139 cgroup version 2 filesystem and cgroup version 1 named hierarchies
140 (i.e., cgroup filesystems mounted with the "none,name=" option).
141 Holding CAP_SYS_ADMIN within the user namespace that owns a process's
143 PID namespace allows (since Linux 3.8) that process to mount /proc
144 filesystems.
145 Note, however, that mounting block-based filesystems can be done only
147 by a process that holds CAP_SYS_ADMIN in the initial user namespace.
148 Interaction of user namespaces and other types of namespaces
150 Starting in Linux 3.8, unprivileged processes can create user name‐
151 spaces, and the other types of namespaces can be created with just the
152 CAP_SYS_ADMIN capability in the caller's user namespace.
153 When a nonuser namespace is created, it is owned by the user namespace
155 in which the creating process was a member at the time of the creation
156 of the namespace. Privileged operations on resources governed by the
157 nonuser namespace require that the process has the necessary capabili‐
158 ties in the user namespace that owns the nonuser namespace.
159 If CLONE_NEWUSER is specified along with other CLONE_NEW* flags in a
161 single clone(2) or unshare(2) call, the user namespace is guaranteed to
162 be created first, giving the child (clone(2)) or caller (unshare(2))
163 privileges over the remaining namespaces created by the call. Thus, it
164 is possible for an unprivileged caller to specify this combination of
165 flags.
166 When a new namespace (other than a user namespace) is created via
168 clone(2) or unshare(2), the kernel records the user namespace of the
169 creating process as the owner of the new namespace. (This association
170 can't be changed.) When a process in the new namespace subsequently
171 performs privileged operations that operate on global resources iso‐
172 lated by the namespace, the permission checks are performed according
173 to the process's capabilities in the user namespace that the kernel as‐
174 sociated with the new namespace. For example, suppose that a process
175 attempts to change the hostname (sethostname(2)), a resource governed
176 by the UTS namespace. In this case, the kernel will determine which
177 user namespace owns the process's UTS namespace, and check whether the
178 process has the required capability (CAP_SYS_ADMIN) in that user name‐
179 space.
180 The NS_GET_USERNS ioctl(2) operation can be used to discover the user
182 namespace that owns a nonuser namespace; see ioctl_ns(2).
183 User and group ID mappings: uid_map and gid_map
185 When a user namespace is created, it starts out without a mapping of
186 user IDs (group IDs) to the parent user namespace. The
187 /proc/pid/uid_map and /proc/pid/gid_map files (available since Linux
188 3.5) expose the mappings for user and group IDs inside the user name‐
189 space for the process pid. These files can be read to view the map‐
190 pings in a user namespace and written to (once) to define the mappings.
191 The description in the following paragraphs explains the details for
193 uid_map; gid_map is exactly the same, but each instance of "user ID" is
194 replaced by "group ID".
195 The uid_map file exposes the mapping of user IDs from the user name‐
197 space of the process pid to the user namespace of the process that
198 opened uid_map (but see a qualification to this point below). In other
199 words, processes that are in different user namespaces will potentially
200 see different values when reading from a particular uid_map file, de‐
201 pending on the user ID mappings for the user namespaces of the reading
202 processes.
203 Each line in the uid_map file specifies a 1-to-1 mapping of a range of
205 contiguous user IDs between two user namespaces. (When a user name‐
206 space is first created, this file is empty.) The specification in each
207 line takes the form of three numbers delimited by white space. The
208 first two numbers specify the starting user ID in each of the two user
209 namespaces. The third number specifies the length of the mapped range.
210 In detail, the fields are interpreted as follows:
211 (1) The start of the range of user IDs in the user namespace of the
213 process pid.
214 (2) The start of the range of user IDs to which the user IDs specified
216 by field one map. How field two is interpreted depends on whether
217 the process that opened uid_map and the process pid are in the
218 same user namespace, as follows:
219 (a) If the two processes are in different user namespaces: field
221 two is the start of a range of user IDs in the user namespace
222 of the process that opened uid_map.
223 (b) If the two processes are in the same user namespace: field
225 two is the start of the range of user IDs in the parent user
226 namespace of the process pid. This case enables the opener
227 of uid_map (the common case here is opening
228 /proc/self/uid_map) to see the mapping of user IDs into the
229 user namespace of the process that created this user name‐
230 space.
231 (3) The length of the range of user IDs that is mapped between the two
233 user namespaces.
234 System calls that return user IDs (group IDs)—for example, getuid(2),
236 getgid(2), and the credential fields in the structure returned by
237 stat(2)—return the user ID (group ID) mapped into the caller's user
238 namespace.
239 When a process accesses a file, its user and group IDs are mapped into
241 the initial user namespace for the purpose of permission checking and
242 assigning IDs when creating a file. When a process retrieves file user
243 and group IDs via stat(2), the IDs are mapped in the opposite direc‐
244 tion, to produce values relative to the process user and group ID map‐
245 pings.
246 The initial user namespace has no parent namespace, but, for consis‐
248 tency, the kernel provides dummy user and group ID mapping files for
249 this namespace. Looking at the uid_map file (gid_map is the same) from
250 a shell in the initial namespace shows:
251 $ cat /proc/$$/uid_map
253 0 0 4294967295
254 This mapping tells us that the range starting at user ID 0 in this
256 namespace maps to a range starting at 0 in the (nonexistent) parent
257 namespace, and the length of the range is the largest 32-bit unsigned
258 integer. This leaves 4294967295 (the 32-bit signed -1 value) unmapped.
259 This is deliberate: (uid_t) -1 is used in several interfaces (e.g., se‐
260 treuid(2)) as a way to specify "no user ID". Leaving (uid_t) -1 un‐
261 mapped and unusable guarantees that there will be no confusion when us‐
262 ing these interfaces.
263 Defining user and group ID mappings: writing to uid_map and gid_map
265 After the creation of a new user namespace, the uid_map file of one of
266 the processes in the namespace may be written to once to define the
267 mapping of user IDs in the new user namespace. An attempt to write
268 more than once to a uid_map file in a user namespace fails with the er‐
269 ror EPERM. Similar rules apply for gid_map files.
270 The lines written to uid_map (gid_map) must conform to the following
272 validity rules:
273 • The three fields must be valid numbers, and the last field must be
275 greater than 0.
276 • Lines are terminated by newline characters.
278 • There is a limit on the number of lines in the file. In Linux 4.14
280 and earlier, this limit was (arbitrarily) set at 5 lines. Since
281 Linux 4.15, the limit is 340 lines. In addition, the number of
282 bytes written to the file must be less than the system page size,
283 and the write must be performed at the start of the file (i.e.,
284 lseek(2) and pwrite(2) can't be used to write to nonzero offsets in
285 the file).
286 • The range of user IDs (group IDs) specified in each line cannot
288 overlap with the ranges in any other lines. In the initial imple‐
289 mentation (Linux 3.8), this requirement was satisfied by a simplis‐
290 tic implementation that imposed the further requirement that the
291 values in both field 1 and field 2 of successive lines must be in
292 ascending numerical order, which prevented some otherwise valid maps
293 from being created. Linux 3.9 and later fix this limitation, allow‐
294 ing any valid set of nonoverlapping maps.
295 • At least one line must be written to the file.
297 Writes that violate the above rules fail with the error EINVAL.
299 In order for a process to write to the /proc/pid/uid_map
301 (/proc/pid/gid_map) file, all of the following permission requirements
302 must be met:
303 • The writing process must have the CAP_SETUID (CAP_SETGID) capability
305 in the user namespace of the process pid.
306 • The writing process must either be in the user namespace of the
308 process pid or be in the parent user namespace of the process pid.
309 • The mapped user IDs (group IDs) must in turn have a mapping in the
311 parent user namespace.
312 • If updating /proc/pid/uid_map to create a mapping that maps UID 0 in
314 the parent namespace, then one of the following must be true:
315 (a) if writing process is in the parent user namespace, then it
317 must have the CAP_SETFCAP capability in that user namespace; or
318 (b) if the writing process is in the child user namespace, then the
320 process that created the user namespace must have had the
321 CAP_SETFCAP capability when the namespace was created.
322 This rule has been in place since Linux 5.12. It eliminates an ear‐
324 lier security bug whereby a UID 0 process that lacks the CAP_SETFCAP
325 capability, which is needed to create a binary with namespaced file
326 capabilities (as described in capabilities(7)), could nevertheless
327 create such a binary, by the following steps:
328 (1) Create a new user namespace with the identity mapping (i.e.,
330 UID 0 in the new user namespace maps to UID 0 in the parent
331 namespace), so that UID 0 in both namespaces is equivalent to
332 the same root user ID.
333 (2) Since the child process has the CAP_SETFCAP capability, it
335 could create a binary with namespaced file capabilities that
336 would then be effective in the parent user namespace (because
337 the root user IDs are the same in the two namespaces).
338 • One of the following two cases applies:
340 (a) Either the writing process has the CAP_SETUID (CAP_SETGID) ca‐
342 pability in the parent user namespace.
343 • No further restrictions apply: the process can make mappings
345 to arbitrary user IDs (group IDs) in the parent user name‐
346 space.
347 (b) Or otherwise all of the following restrictions apply:
349 • The data written to uid_map (gid_map) must consist of a sin‐
351 gle line that maps the writing process's effective user ID
352 (group ID) in the parent user namespace to a user ID (group
353 ID) in the user namespace.
354 • The writing process must have the same effective user ID as
356 the process that created the user namespace.
357 • In the case of gid_map, use of the setgroups(2) system call
359 must first be denied by writing "deny" to the /proc/pid/set‐
360 groups file (see below) before writing to gid_map.
361 Writes that violate the above rules fail with the error EPERM.
363 Project ID mappings: projid_map
365 Similarly to user and group ID mappings, it is possible to create
366 project ID mappings for a user namespace. (Project IDs are used for
367 disk quotas; see setquota(8) and quotactl(2).)
368 Project ID mappings are defined by writing to the /proc/pid/projid_map
370 file (present since Linux 3.7).
371 The validity rules for writing to the /proc/pid/projid_map file are as
373 for writing to the uid_map file; violation of these rules causes
374 write(2) to fail with the error EINVAL.
375 The permission rules for writing to the /proc/pid/projid_map file are
377 as follows:
378 • The writing process must either be in the user namespace of the
380 process pid or be in the parent user namespace of the process pid.
381 • The mapped project IDs must in turn have a mapping in the parent
383 user namespace.
384 Violation of these rules causes write(2) to fail with the error EPERM.
386 Interaction with system calls that change process UIDs or GIDs
388 In a user namespace where the uid_map file has not been written, the
389 system calls that change user IDs will fail. Similarly, if the gid_map
390 file has not been written, the system calls that change group IDs will
391 fail. After the uid_map and gid_map files have been written, only the
392 mapped values may be used in system calls that change user and group
393 IDs.
394 For user IDs, the relevant system calls include setuid(2), setfsuid(2),
396 setreuid(2), and setresuid(2). For group IDs, the relevant system
397 calls include setgid(2), setfsgid(2), setregid(2), setresgid(2), and
398 setgroups(2).
399 Writing "deny" to the /proc/pid/setgroups file before writing to
401 /proc/pid/gid_map will permanently disable setgroups(2) in a user name‐
402 space and allow writing to /proc/pid/gid_map without having the
403 CAP_SETGID capability in the parent user namespace.
404 The /proc/pid/setgroups file
406 The /proc/pid/setgroups file displays the string "allow" if processes
407 in the user namespace that contains the process pid are permitted to
408 employ the setgroups(2) system call; it displays "deny" if setgroups(2)
409 is not permitted in that user namespace. Note that regardless of the
410 value in the /proc/pid/setgroups file (and regardless of the process's
411 capabilities), calls to setgroups(2) are also not permitted if
412 /proc/pid/gid_map has not yet been set.
413 A privileged process (one with the CAP_SYS_ADMIN capability in the
415 namespace) may write either of the strings "allow" or "deny" to this
416 file before writing a group ID mapping for this user namespace to the
417 file /proc/pid/gid_map. Writing the string "deny" prevents any process
418 in the user namespace from employing setgroups(2).
419 The essence of the restrictions described in the preceding paragraph is
421 that it is permitted to write to /proc/pid/setgroups only so long as
422 calling setgroups(2) is disallowed because /proc/pid/gid_map has not
423 been set. This ensures that a process cannot transition from a state
424 where setgroups(2) is allowed to a state where setgroups(2) is denied;
425 a process can transition only from setgroups(2) being disallowed to
426 setgroups(2) being allowed.
427 The default value of this file in the initial user namespace is "al‐
429 low".
430 Once /proc/pid/gid_map has been written to (which has the effect of en‐
432 abling setgroups(2) in the user namespace), it is no longer possible to
433 disallow setgroups(2) by writing "deny" to /proc/pid/setgroups (the
434 write fails with the error EPERM).
435 A child user namespace inherits the /proc/pid/setgroups setting from
437 its parent.
438 If the setgroups file has the value "deny", then the setgroups(2) sys‐
440 tem call can't subsequently be reenabled (by writing "allow" to the
441 file) in this user namespace. (Attempts to do so fail with the error
442 EPERM.) This restriction also propagates down to all child user name‐
443 spaces of this user namespace.
444 The /proc/pid/setgroups file was added in Linux 3.19, but was back‐
446 ported to many earlier stable kernel series, because it addresses a se‐
447 curity issue. The issue concerned files with permissions such as
448 "rwx---rwx". Such files give fewer permissions to "group" than they do
449 to "other". This means that dropping groups using setgroups(2) might
450 allow a process file access that it did not formerly have. Before the
451 existence of user namespaces this was not a concern, since only a priv‐
452 ileged process (one with the CAP_SETGID capability) could call set‐
453 groups(2). However, with the introduction of user namespaces, it be‐
454 came possible for an unprivileged process to create a new namespace in
455 which the user had all privileges. This then allowed formerly unprivi‐
456 leged users to drop groups and thus gain file access that they did not
457 previously have. The /proc/pid/setgroups file was added to address
458 this security issue, by denying any pathway for an unprivileged process
459 to drop groups with setgroups(2).
460 Unmapped user and group IDs
462 There are various places where an unmapped user ID (group ID) may be
463 exposed to user space. For example, the first process in a new user
464 namespace may call getuid(2) before a user ID mapping has been defined
465 for the namespace. In most such cases, an unmapped user ID is con‐
466 verted to the overflow user ID (group ID); the default value for the
467 overflow user ID (group ID) is 65534. See the descriptions of
468 /proc/sys/kernel/overflowuid and /proc/sys/kernel/overflowgid in
469 proc(5).
470 The cases where unmapped IDs are mapped in this fashion include system
472 calls that return user IDs (getuid(2), getgid(2), and similar), creden‐
473 tials passed over a UNIX domain socket, credentials returned by
474 stat(2), waitid(2), and the System V IPC "ctl" IPC_STAT operations,
475 credentials exposed by /proc/pid/status and the files in
476 /proc/sysvipc/*, credentials returned via the si_uid field in the sig‐
477 info_t received with a signal (see sigaction(2)), credentials written
478 to the process accounting file (see acct(5)), and credentials returned
479 with POSIX message queue notifications (see mq_notify(3)).
480 There is one notable case where unmapped user and group IDs are not
482 converted to the corresponding overflow ID value. When viewing a
483 uid_map or gid_map file in which there is no mapping for the second
484 field, that field is displayed as 4294967295 (-1 as an unsigned inte‐
485 ger).
486 Accessing files
488 In order to determine permissions when an unprivileged process accesses
489 a file, the process credentials (UID, GID) and the file credentials are
490 in effect mapped back to what they would be in the initial user name‐
491 space and then compared to determine the permissions that the process
492 has on the file. The same is also true of other objects that employ
493 the credentials plus permissions mask accessibility model, such as Sys‐
494 tem V IPC objects.
495 Operation of file-related capabilities
497 Certain capabilities allow a process to bypass various kernel-enforced
498 restrictions when performing operations on files owned by other users
499 or groups. These capabilities are: CAP_CHOWN, CAP_DAC_OVERRIDE,
500 CAP_DAC_READ_SEARCH, CAP_FOWNER, and CAP_FSETID.
501 Within a user namespace, these capabilities allow a process to bypass
503 the rules if the process has the relevant capability over the file,
504 meaning that:
505 • the process has the relevant effective capability in its user name‐
507 space; and
508 • the file's user ID and group ID both have valid mappings in the user
510 namespace.
511 The CAP_FOWNER capability is treated somewhat exceptionally: it allows
513 a process to bypass the corresponding rules so long as at least the
514 file's user ID has a mapping in the user namespace (i.e., the file's
515 group ID does not need to have a valid mapping).
516 Set-user-ID and set-group-ID programs
518 When a process inside a user namespace executes a set-user-ID (set-
519 group-ID) program, the process's effective user (group) ID inside the
520 namespace is changed to whatever value is mapped for the user (group)
521 ID of the file. However, if either the user or the group ID of the
522 file has no mapping inside the namespace, the set-user-ID (set-group-
523 ID) bit is silently ignored: the new program is executed, but the
524 process's effective user (group) ID is left unchanged. (This mirrors
525 the semantics of executing a set-user-ID or set-group-ID program that
526 resides on a filesystem that was mounted with the MS_NOSUID flag, as
527 described in mount(2).)
528 Miscellaneous
530 When a process's user and group IDs are passed over a UNIX domain
531 socket to a process in a different user namespace (see the description
532 of SCM_CREDENTIALS in unix(7)), they are translated into the corre‐
533 sponding values as per the receiving process's user and group ID map‐
534 pings.
535
537 Linux.
538
540 Over the years, there have been a lot of features that have been added
541 to the Linux kernel that have been made available only to privileged
542 users because of their potential to confuse set-user-ID-root applica‐
543 tions. In general, it becomes safe to allow the root user in a user
544 namespace to use those features because it is impossible, while in a
545 user namespace, to gain more privilege than the root user of a user
546 namespace has.
547 Global root
549 The term "global root" is sometimes used as a shorthand for user ID 0
550 in the initial user namespace.
551 Availability
553 Use of user namespaces requires a kernel that is configured with the
554 CONFIG_USER_NS option. User namespaces require support in a range of
555 subsystems across the kernel. When an unsupported subsystem is config‐
556 ured into the kernel, it is not possible to configure user namespaces
557 support.
558 As at Linux 3.8, most relevant subsystems supported user namespaces,
560 but a number of filesystems did not have the infrastructure needed to
561 map user and group IDs between user namespaces. Linux 3.9 added the
562 required infrastructure support for many of the remaining unsupported
563 filesystems (Plan 9 (9P), Andrew File System (AFS), Ceph, CIFS, CODA,
564 NFS, and OCFS2). Linux 3.12 added support for the last of the unsup‐
565 ported major filesystems, XFS.
566
568 The program below is designed to allow experimenting with user name‐
569 spaces, as well as other types of namespaces. It creates namespaces as
570 specified by command-line options and then executes a command inside
571 those namespaces. The comments and usage() function inside the program
572 provide a full explanation of the program. The following shell session
573 demonstrates its use.
574 First, we look at the run-time environment:
576 $ uname -rs # Need Linux 3.8 or later
578 Linux 3.8.0
579 $ id -u # Running as unprivileged user
580 1000
581 $ id -g
582 1000
583 Now start a new shell in new user (-U), mount (-m), and PID (-p) name‐
585 spaces, with user ID (-M) and group ID (-G) 1000 mapped to 0 inside the
586 user namespace:
587 $ ./userns_child_exec -p -m -U -M '0 1000 1' -G '0 1000 1' bash
589 The shell has PID 1, because it is the first process in the new PID
591 namespace:
592 bash$ echo $$
594 1
595 Mounting a new /proc filesystem and listing all of the processes visi‐
597 ble in the new PID namespace shows that the shell can't see any pro‐
598 cesses outside the PID namespace:
599 bash$ mount -t proc proc /proc
601 bash$ ps ax
602 PID TTY STAT TIME COMMAND
603 1 pts/3 S 0:00 bash
604 22 pts/3 R+ 0:00 ps ax
605 Inside the user namespace, the shell has user and group ID 0, and a
607 full set of permitted and effective capabilities:
608 bash$ cat /proc/$$/status | egrep '^[UG]id'
610 Uid: 0 0 0 0
611 Gid: 0 0 0 0
612 bash$ cat /proc/$$/status | egrep '^Cap(Prm|Inh|Eff)'
613 CapInh: 0000000000000000
614 CapPrm: 0000001fffffffff
615 CapEff: 0000001fffffffff
616 Program source
618 /* userns_child_exec.c
620 Licensed under GNU General Public License v2 or later
622 Create a child process that executes a shell command in new
624 namespace(s); allow UID and GID mappings to be specified when
625 creating a user namespace.
626 */
627 #define _GNU_SOURCE
628 #include <err.h>
629 #include <sched.h>
630 #include <unistd.h>
631 #include <stdint.h>
632 #include <stdlib.h>
633 #include <sys/wait.h>
634 #include <signal.h>
635 #include <fcntl.h>
636 #include <stdio.h>
637 #include <string.h>
638 #include <limits.h>
639 #include <errno.h>
640 struct child_args {
642 char **argv; /* Command to be executed by child, with args */
643 int pipe_fd[2]; /* Pipe used to synchronize parent and child */
644 };
645 static int verbose;
647 static void
649 usage(char *pname)
650 {
651 fprintf(stderr, "Usage: %s [options] cmd [arg...]\n\n", pname);
652 fprintf(stderr, "Create a child process that executes a shell "
653 "command in a new user namespace,\n"
654 "and possibly also other new namespace(s).\n\n");
655 fprintf(stderr, "Options can be:\n\n");
656 #define fpe(str) fprintf(stderr, " %s", str);
657 fpe("-i New IPC namespace\n");
658 fpe("-m New mount namespace\n");
659 fpe("-n New network namespace\n");
660 fpe("-p New PID namespace\n");
661 fpe("-u New UTS namespace\n");
662 fpe("-U New user namespace\n");
663 fpe("-M uid_map Specify UID map for user namespace\n");
664 fpe("-G gid_map Specify GID map for user namespace\n");
665 fpe("-z Map user's UID and GID to 0 in user namespace\n");
666 fpe(" (equivalent to: -M '0 <uid> 1' -G '0 <gid> 1')\n");
667 fpe("-v Display verbose messages\n");
668 fpe("\n");
669 fpe("If -z, -M, or -G is specified, -U is required.\n");
670 fpe("It is not permitted to specify both -z and either -M or -G.\n");
671 fpe("\n");
672 fpe("Map strings for -M and -G consist of records of the form:\n");
673 fpe("\n");
674 fpe(" ID-inside-ns ID-outside-ns len\n");
675 fpe("\n");
676 fpe("A map string can contain multiple records, separated"
677 " by commas;\n");
678 fpe("the commas are replaced by newlines before writing"
679 " to map files.\n");
680 exit(EXIT_FAILURE);
682 }
683 /* Update the mapping file 'map_file', with the value provided in
685 'mapping', a string that defines a UID or GID mapping. A UID or
686 GID mapping consists of one or more newline-delimited records
687 of the form:
688 ID_inside-ns ID-outside-ns length
690 Requiring the user to supply a string that contains newlines is
692 of course inconvenient for command-line use. Thus, we permit the
693 use of commas to delimit records in this string, and replace them
694 with newlines before writing the string to the file. */
695 static void
697 update_map(char *mapping, char *map_file)
698 {
699 int fd;
700 size_t map_len; /* Length of 'mapping' */
701 /* Replace commas in mapping string with newlines. */
703 map_len = strlen(mapping);
705 for (size_t j = 0; j < map_len; j++)
706 if (mapping[j] == ',')
707 mapping[j] = '\n';
708 fd = open(map_file, O_RDWR);
710 if (fd == -1) {
711 fprintf(stderr, "ERROR: open %s: %s\n", map_file,
712 strerror(errno));
713 exit(EXIT_FAILURE);
714 }
715 if (write(fd, mapping, map_len) != map_len) {
717 fprintf(stderr, "ERROR: write %s: %s\n", map_file,
718 strerror(errno));
719 exit(EXIT_FAILURE);
720 }
721 close(fd);
723 }
724 /* Linux 3.19 made a change in the handling of setgroups(2) and
726 the 'gid_map' file to address a security issue. The issue
727 allowed *unprivileged* users to employ user namespaces in
728 order to drop groups. The upshot of the 3.19 changes is that
729 in order to update the 'gid_maps' file, use of the setgroups()
730 system call in this user namespace must first be disabled by
731 writing "deny" to one of the /proc/PID/setgroups files for
732 this namespace. That is the purpose of the following function. */
733 static void
735 proc_setgroups_write(pid_t child_pid, char *str)
736 {
737 char setgroups_path[PATH_MAX];
738 int fd;
739 snprintf(setgroups_path, PATH_MAX, "/proc/%jd/setgroups",
741 (intmax_t) child_pid);
742 fd = open(setgroups_path, O_RDWR);
744 if (fd == -1) {
745 /* We may be on a system that doesn't support
747 /proc/PID/setgroups. In that case, the file won't exist,
748 and the system won't impose the restrictions that Linux 3.19
749 added. That's fine: we don't need to do anything in order
750 to permit 'gid_map' to be updated.
751 However, if the error from open() was something other than
753 the ENOENT error that is expected for that case, let the
754 user know. */
755 if (errno != ENOENT)
757 fprintf(stderr, "ERROR: open %s: %s\n", setgroups_path,
758 strerror(errno));
759 return;
760 }
761 if (write(fd, str, strlen(str)) == -1)
763 fprintf(stderr, "ERROR: write %s: %s\n", setgroups_path,
764 strerror(errno));
765 close(fd);
767 }
768 static int /* Start function for cloned child */
770 childFunc(void *arg)
771 {
772 struct child_args *args = arg;
773 char ch;
774 /* Wait until the parent has updated the UID and GID mappings.
776 See the comment in main(). We wait for end of file on a
777 pipe that will be closed by the parent process once it has
778 updated the mappings. */
779 close(args->pipe_fd[1]); /* Close our descriptor for the write
781 end of the pipe so that we see EOF
782 when parent closes its descriptor. */
783 if (read(args->pipe_fd[0], &ch, 1) != 0) {
784 fprintf(stderr,
785 "Failure in child: read from pipe returned != 0\n");
786 exit(EXIT_FAILURE);
787 }
788 close(args->pipe_fd[0]);
790 /* Execute a shell command. */
792 printf("About to exec %s\n", args->argv[0]);
794 execvp(args->argv[0], args->argv);
795 err(EXIT_FAILURE, "execvp");
796 }
797 #define STACK_SIZE (1024 * 1024)
799 static char child_stack[STACK_SIZE]; /* Space for child's stack */
801 int
803 main(int argc, char *argv[])
804 {
805 int flags, opt, map_zero;
806 pid_t child_pid;
807 struct child_args args;
808 char *uid_map, *gid_map;
809 const int MAP_BUF_SIZE = 100;
810 char map_buf[MAP_BUF_SIZE];
811 char map_path[PATH_MAX];
812 /* Parse command-line options. The initial '+' character in
814 the final getopt() argument prevents GNU-style permutation
815 of command-line options. That's useful, since sometimes
816 the 'command' to be executed by this program itself
817 has command-line options. We don't want getopt() to treat
818 those as options to this program. */
819 flags = 0;
821 verbose = 0;
822 gid_map = NULL;
823 uid_map = NULL;
824 map_zero = 0;
825 while ((opt = getopt(argc, argv, "+imnpuUM:G:zv")) != -1) {
826 switch (opt) {
827 case 'i': flags |= CLONE_NEWIPC; break;
828 case 'm': flags |= CLONE_NEWNS; break;
829 case 'n': flags |= CLONE_NEWNET; break;
830 case 'p': flags |= CLONE_NEWPID; break;
831 case 'u': flags |= CLONE_NEWUTS; break;
832 case 'v': verbose = 1; break;
833 case 'z': map_zero = 1; break;
834 case 'M': uid_map = optarg; break;
835 case 'G': gid_map = optarg; break;
836 case 'U': flags |= CLONE_NEWUSER; break;
837 default: usage(argv[0]);
838 }
839 }
840 /* -M or -G without -U is nonsensical */
842 if (((uid_map != NULL || gid_map != NULL || map_zero) &&
844 !(flags & CLONE_NEWUSER)) ||
845 (map_zero && (uid_map != NULL || gid_map != NULL)))
846 usage(argv[0]);
847 args.argv = &argv[optind];
849 /* We use a pipe to synchronize the parent and child, in order to
851 ensure that the parent sets the UID and GID maps before the child
852 calls execve(). This ensures that the child maintains its
853 capabilities during the execve() in the common case where we
854 want to map the child's effective user ID to 0 in the new user
855 namespace. Without this synchronization, the child would lose
856 its capabilities if it performed an execve() with nonzero
857 user IDs (see the capabilities(7) man page for details of the
858 transformation of a process's capabilities during execve()). */
859 if (pipe(args.pipe_fd) == -1)
861 err(EXIT_FAILURE, "pipe");
862 /* Create the child in new namespace(s). */
864 child_pid = clone(childFunc, child_stack + STACK_SIZE,
866 flags | SIGCHLD, &args);
867 if (child_pid == -1)
868 err(EXIT_FAILURE, "clone");
869 /* Parent falls through to here. */
871 if (verbose)
873 printf("%s: PID of child created by clone() is %jd\n",
874 argv[0], (intmax_t) child_pid);
875 /* Update the UID and GID maps in the child. */
877 if (uid_map != NULL || map_zero) {
879 snprintf(map_path, PATH_MAX, "/proc/%jd/uid_map",
880 (intmax_t) child_pid);
881 if (map_zero) {
882 snprintf(map_buf, MAP_BUF_SIZE, "0 %jd 1",
883 (intmax_t) getuid());
884 uid_map = map_buf;
885 }
886 update_map(uid_map, map_path);
887 }
888 if (gid_map != NULL || map_zero) {
890 proc_setgroups_write(child_pid, "deny");
891 snprintf(map_path, PATH_MAX, "/proc/%jd/gid_map",
893 (intmax_t) child_pid);
894 if (map_zero) {
895 snprintf(map_buf, MAP_BUF_SIZE, "0 %ld 1",
896 (intmax_t) getgid());
897 gid_map = map_buf;
898 }
899 update_map(gid_map, map_path);
900 }
901 /* Close the write end of the pipe, to signal to the child that we
903 have updated the UID and GID maps. */
904 close(args.pipe_fd[1]);
906 if (waitpid(child_pid, NULL, 0) == -1) /* Wait for child */
908 err(EXIT_FAILURE, "waitpid");
909 if (verbose)
911 printf("%s: terminating\n", argv[0]);
912 exit(EXIT_SUCCESS);
914 }
915
917 newgidmap(1), newuidmap(1), clone(2), ptrace(2), setns(2), unshare(2),
918 proc(5), subgid(5), subuid(5), capabilities(7), cgroup_namespaces(7),
919 credentials(7), namespaces(7), pid_namespaces(7)
920 The kernel source file Documentation/admin-guide/namespaces/re‐
922 source-control.rst.
923Linux man-pages 6.05 2023-05-03 user_namespaces(7)