| 1 |
.\" Copyright 2013-2014, Michael Kerrisk <[email protected]> |
| 2 |
.\" Copyright 2012-2014, Eric W. Biederman <[email protected]> |
| 3 |
.\" Copyright, the authors of the Linux man-pages project |
| 4 |
.\" |
| 5 |
.\" SPDX-License-Identifier: Linux-man-pages-copyleft |
| 6 |
.\" |
| 7 |
.TH user_namespaces 7 2026-02-08 "Linux man-pages 6.18" |
| 8 |
.SH NAME |
| 9 |
user_namespaces \- overview of Linux user namespaces |
| 10 |
.SH DESCRIPTION |
| 11 |
For an overview of namespaces, see |
| 12 |
.BR namespaces (7). |
| 13 |
.P |
| 14 |
User namespaces isolate security-related identifiers and attributes, |
| 15 |
in particular, |
| 16 |
user IDs and group IDs |
| 17 |
(see |
| 18 |
.BR credentials (7)), |
| 19 |
the root directory, |
| 20 |
keys |
| 21 |
(see |
| 22 |
.BR keyrings (7)), |
| 23 |
.\" FIXME: This page says very little about the interaction |
| 24 |
.\" of user namespaces and keys. Add something on this topic. |
| 25 |
and capabilities (see |
| 26 |
.BR capabilities (7)). |
| 27 |
A process's user and group IDs can be different |
| 28 |
inside and outside a user namespace. |
| 29 |
In particular, |
| 30 |
a process can have a normal unprivileged user ID outside a user namespace |
| 31 |
while at the same time having a user ID of 0 inside the namespace; |
| 32 |
in other words, |
| 33 |
the process has full privileges for operations inside the user namespace, |
| 34 |
but is unprivileged for operations outside the namespace. |
| 35 |
.\" |
| 36 |
.\" ============================================================ |
| 37 |
.\" |
| 38 |
.SS Nested namespaces, namespace membership |
| 39 |
User namespaces can be nested; |
| 40 |
that is, |
| 41 |
each user namespace |
| 42 |
\[em]except the initial ("root") namespace\[em] |
| 43 |
has a parent user namespace, |
| 44 |
and can have zero or more child user namespaces. |
| 45 |
The parent user namespace is |
| 46 |
the user namespace of the process that creates the user namespace |
| 47 |
via a call to |
| 48 |
.BR unshare (2) |
| 49 |
or |
| 50 |
.BR clone (2) |
| 51 |
with the |
| 52 |
.B CLONE_NEWUSER |
| 53 |
flag. |
| 54 |
.P |
| 55 |
.\" commit 8742f229b635bf1c1c84a3dfe5e47c814c20b5c8 |
| 56 |
The kernel imposes (since Linux 3.11) |
| 57 |
a limit of 32 nested levels of user namespaces. |
| 58 |
.\" FIXME Explain the rationale for this limit. (What is the rationale?) |
| 59 |
Calls to |
| 60 |
.BR unshare (2) |
| 61 |
or |
| 62 |
.BR clone (2) |
| 63 |
that would cause this limit to be exceeded fail with the error |
| 64 |
.BR EUSERS . |
| 65 |
.P |
| 66 |
Each process is a member of exactly one user namespace. |
| 67 |
A process created via |
| 68 |
.BR fork (2) |
| 69 |
or |
| 70 |
.BR clone (2) |
| 71 |
without the |
| 72 |
.B CLONE_NEWUSER |
| 73 |
flag is a member of the same user namespace as its parent. |
| 74 |
A single-threaded process can join another user namespace with |
| 75 |
.BR setns (2) |
| 76 |
if it has the |
| 77 |
.B CAP_SYS_ADMIN |
| 78 |
in that namespace; |
| 79 |
upon doing so, |
| 80 |
it gains a full set of capabilities in that namespace. |
| 81 |
.P |
| 82 |
A call to |
| 83 |
.BR clone (2) |
| 84 |
or |
| 85 |
.BR unshare (2) |
| 86 |
with the |
| 87 |
.B CLONE_NEWUSER |
| 88 |
flag makes the new child process |
| 89 |
(for |
| 90 |
.BR clone (2)) |
| 91 |
or the caller |
| 92 |
(for |
| 93 |
.BR unshare (2)) |
| 94 |
a member of the new user namespace created by the call. |
| 95 |
.P |
| 96 |
The |
| 97 |
.B NS_GET_PARENT |
| 98 |
.BR ioctl (2) |
| 99 |
operation can be used |
| 100 |
to discover the parental relationship between user namespaces; |
| 101 |
see |
| 102 |
.BR ioctl_nsfs (2). |
| 103 |
.P |
| 104 |
A task that changes one of its effective IDs |
| 105 |
will have its dumpability reset to the value in |
| 106 |
.IR /proc/sys/fs/suid_dumpable . |
| 107 |
This may affect the ownership of proc files of child processes |
| 108 |
and may thus cause the parent to |
| 109 |
lack the permissions |
| 110 |
to write to mapping files of child processes running in a new user namespace. |
| 111 |
In such cases, |
| 112 |
making the parent process dumpable, |
| 113 |
using |
| 114 |
.B PR_SET_DUMPABLE |
| 115 |
in a call to |
| 116 |
.BR prctl (2), |
| 117 |
before creating a child process in a new user namespace |
| 118 |
may rectify this problem. |
| 119 |
See |
| 120 |
.BR prctl (2) |
| 121 |
and |
| 122 |
.BR proc (5) |
| 123 |
for details on how ownership is affected. |
| 124 |
.\" |
| 125 |
.\" ============================================================ |
| 126 |
.\" |
| 127 |
.SS Capabilities |
| 128 |
The child process created by |
| 129 |
.BR clone (2) |
| 130 |
with the |
| 131 |
.B CLONE_NEWUSER |
| 132 |
flag starts out with a complete set of capabilities |
| 133 |
in the new user namespace. |
| 134 |
Likewise, |
| 135 |
a process that creates a new user namespace using |
| 136 |
.BR unshare (2) |
| 137 |
or joins an existing user namespace using |
| 138 |
.BR setns (2) |
| 139 |
gains a full set of capabilities in that namespace. |
| 140 |
On the other hand, |
| 141 |
that process has no capabilities in the parent |
| 142 |
(in the case of |
| 143 |
.BR clone (2)) |
| 144 |
or previous |
| 145 |
(in the case of |
| 146 |
.BR unshare (2) |
| 147 |
and |
| 148 |
.BR setns (2)) |
| 149 |
user namespace, |
| 150 |
even if the new namespace is created or joined by the root user |
| 151 |
(i.e., |
| 152 |
a process with user ID 0 in the root namespace). |
| 153 |
.P |
| 154 |
Note that a call to |
| 155 |
.BR execve (2) |
| 156 |
will cause a process's capabilities to be recalculated in the usual way |
| 157 |
(see |
| 158 |
.BR capabilities (7)). |
| 159 |
Consequently, |
| 160 |
unless the process has a user ID of 0 within the namespace, |
| 161 |
or the executable file has a nonempty inheritable capabilities mask, |
| 162 |
the process will lose all capabilities. |
| 163 |
See the discussion of user and group ID mappings, below. |
| 164 |
.P |
| 165 |
A call to |
| 166 |
.BR clone (2) |
| 167 |
or |
| 168 |
.BR unshare (2) |
| 169 |
using the |
| 170 |
.B CLONE_NEWUSER |
| 171 |
flag |
| 172 |
or a call to |
| 173 |
.BR setns (2) |
| 174 |
that moves the caller into another user namespace |
| 175 |
sets the "securebits" flags |
| 176 |
(see |
| 177 |
.BR capabilities (7)) |
| 178 |
to their default values (all flags disabled) |
| 179 |
in the child |
| 180 |
(for |
| 181 |
.BR clone (2)) |
| 182 |
or caller |
| 183 |
(for |
| 184 |
.BR unshare (2) |
| 185 |
or |
| 186 |
.BR setns (2)). |
| 187 |
Note that because the caller no longer has capabilities |
| 188 |
in its original user namespace |
| 189 |
after a call to |
| 190 |
.BR setns (2), |
| 191 |
it is not possible for a process to reset its "securebits" flags |
| 192 |
while retaining its user namespace membership |
| 193 |
by using a pair of |
| 194 |
.BR setns (2) |
| 195 |
calls to move to another user namespace |
| 196 |
and then return to its original user namespace. |
| 197 |
.P |
| 198 |
The rules for determining |
| 199 |
whether or not a process has a capability |
| 200 |
in a particular user namespace |
| 201 |
are as follows: |
| 202 |
.IP \[bu] 3 |
| 203 |
A process has a capability inside a user namespace |
| 204 |
if it is a member of that namespace |
| 205 |
and it has the capability in its effective capability set. |
| 206 |
A process can gain capabilities in its effective capability set |
| 207 |
in various ways. |
| 208 |
For example, |
| 209 |
it may execute a set-user-ID program |
| 210 |
or an executable with associated file capabilities. |
| 211 |
In addition, |
| 212 |
a process may gain capabilities via the effect of |
| 213 |
.BR clone (2), |
| 214 |
.BR unshare (2), |
| 215 |
or |
| 216 |
.BR setns (2), |
| 217 |
as already described. |
| 218 |
.\" In the 3.8 sources, see security/commoncap.c::cap_capable(): |
| 219 |
.IP \[bu] |
| 220 |
If a process has a capability in a user namespace, |
| 221 |
then it has that capability |
| 222 |
in all child (and further removed descendant) namespaces as well. |
| 223 |
.IP \[bu] |
| 224 |
.\" * The owner of the user namespace in the parent of the |
| 225 |
.\" * user namespace has all caps. |
| 226 |
When a user namespace is created, |
| 227 |
the kernel records the effective user ID of the creating process |
| 228 |
as being the "owner" of the namespace. |
| 229 |
.\" (and likewise associates the effective group ID of the creating process |
| 230 |
.\" with the namespace). |
| 231 |
A process that resides |
| 232 |
in the parent of the user namespace |
| 233 |
.\" See kernel commit 520d9eabce18edfef76a60b7b839d54facafe1f9 for a fix |
| 234 |
.\" on this point |
| 235 |
and whose effective user ID matches the owner of the namespace |
| 236 |
has all capabilities in the namespace. |
| 237 |
.\" This includes the case where the process executes a set-user-ID |
| 238 |
.\" program that confers the effective UID of the creator of the namespace. |
| 239 |
By virtue of the previous rule, |
| 240 |
this means that the process has all capabilities |
| 241 |
in all further removed descendant user namespaces as well. |
| 242 |
The |
| 243 |
.B NS_GET_OWNER_UID |
| 244 |
.BR ioctl (2) |
| 245 |
operation can be used to discover the user ID of the owner of the namespace; |
| 246 |
see |
| 247 |
.BR ioctl_nsfs (2). |
| 248 |
.\" |
| 249 |
.\" ============================================================ |
| 250 |
.\" |
| 251 |
.SS Effect of capabilities within a user namespace |
| 252 |
Having a capability inside a user namespace |
| 253 |
permits a process to perform operations (that require privilege) |
| 254 |
only on resources governed by that namespace. |
| 255 |
In other words, |
| 256 |
having a capability in a user namespace permits a process |
| 257 |
to perform privileged operations |
| 258 |
on resources that are governed by |
| 259 |
(nonuser) namespaces owned by (associated with) the user namespace |
| 260 |
(see the next subsection). |
| 261 |
.P |
| 262 |
On the other hand, |
| 263 |
there are many privileged operations that affect |
| 264 |
resources that are not associated with any namespace type, |
| 265 |
for example, |
| 266 |
changing the system (i.e., calendar) time |
| 267 |
(governed by |
| 268 |
.BR CAP_SYS_TIME ), |
| 269 |
loading a kernel module |
| 270 |
(governed by |
| 271 |
.BR CAP_SYS_MODULE ), |
| 272 |
and creating a device |
| 273 |
(governed by |
| 274 |
.BR CAP_MKNOD ). |
| 275 |
Only a process with privileges in the |
| 276 |
.I initial |
| 277 |
user namespace can perform such operations. |
| 278 |
.P |
| 279 |
Holding |
| 280 |
.B CAP_SYS_ADMIN |
| 281 |
within the user namespace that owns a process's mount namespace |
| 282 |
allows that process to create bind mounts |
| 283 |
and mount the following types of filesystems: |
| 284 |
.\" fs_flags = FS_USERNS_MOUNT in kernel sources |
| 285 |
.P |
| 286 |
.RS 4 |
| 287 |
.PD 0 |
| 288 |
.IP \[bu] 3 |
| 289 |
.I /proc |
| 290 |
(since Linux 3.8) |
| 291 |
.IP \[bu] |
| 292 |
.I /sys |
| 293 |
(since Linux 3.8) |
| 294 |
.IP \[bu] |
| 295 |
.I devpts |
| 296 |
(since Linux 3.9) |
| 297 |
.IP \[bu] |
| 298 |
.BR tmpfs (5) |
| 299 |
(since Linux 3.9) |
| 300 |
.IP \[bu] |
| 301 |
.I ramfs |
| 302 |
(since Linux 3.9) |
| 303 |
.IP \[bu] |
| 304 |
.I mqueue |
| 305 |
(since Linux 3.9) |
| 306 |
.IP \[bu] |
| 307 |
.I bpf |
| 308 |
.\" commit b2197755b2633e164a439682fb05a9b5ea48f706 |
| 309 |
(since Linux 4.4) |
| 310 |
.IP \[bu] |
| 311 |
.I overlayfs |
| 312 |
.\" commit 92dbc9dedccb9759c7f9f2f0ae6242396376988f |
| 313 |
.\" commit 4cb2c00c43b3fe88b32f29df4f76da1b92c33224 |
| 314 |
(since Linux 5.11) |
| 315 |
.PD |
| 316 |
.RE |
| 317 |
.P |
| 318 |
Holding |
| 319 |
.B CAP_SYS_ADMIN |
| 320 |
within the user namespace that owns a process's cgroup namespace |
| 321 |
allows (since Linux 4.6) |
| 322 |
that process to mount |
| 323 |
the cgroup version 2 filesystem |
| 324 |
and cgroup version 1 named hierarchies |
| 325 |
(i.e., cgroup filesystems mounted with the |
| 326 |
.I \[dq]none,name=\[dq] |
| 327 |
option). |
| 328 |
.P |
| 329 |
Holding |
| 330 |
.B CAP_SYS_ADMIN |
| 331 |
within the user namespace that owns a process's PID namespace |
| 332 |
allows (since Linux 3.8) |
| 333 |
that process to mount |
| 334 |
.I /proc |
| 335 |
filesystems. |
| 336 |
.P |
| 337 |
However, |
| 338 |
mounting block-based filesystems can be done |
| 339 |
only by a process that holds |
| 340 |
.B CAP_SYS_ADMIN |
| 341 |
in the initial user namespace. |
| 342 |
.\" |
| 343 |
.\" ============================================================ |
| 344 |
.\" |
| 345 |
.SS Interaction of user namespaces and other types of namespaces |
| 346 |
Since Linux 3.8, |
| 347 |
unprivileged processes can create user namespaces, |
| 348 |
and the other types of namespaces can be created with just the |
| 349 |
.B CAP_SYS_ADMIN |
| 350 |
capability in the caller's user namespace. |
| 351 |
.P |
| 352 |
When a nonuser namespace is created, |
| 353 |
it is owned by the user namespace in which the creating process |
| 354 |
was a member at the time of the creation of the namespace. |
| 355 |
Privileged operations on resources governed by the nonuser namespace |
| 356 |
require that the process has the necessary capabilities |
| 357 |
in the user namespace that owns the nonuser namespace. |
| 358 |
.P |
| 359 |
If |
| 360 |
.B CLONE_NEWUSER |
| 361 |
is specified along with other |
| 362 |
.B CLONE_NEW* |
| 363 |
flags in a single |
| 364 |
.BR clone (2) |
| 365 |
or |
| 366 |
.BR unshare (2) |
| 367 |
call, |
| 368 |
the user namespace is guaranteed to be created first, |
| 369 |
giving the child |
| 370 |
.RB ( clone (2)) |
| 371 |
or caller |
| 372 |
.RB ( unshare (2)) |
| 373 |
privileges over the remaining namespaces created by the call. |
| 374 |
Thus, |
| 375 |
it is possible for an unprivileged caller to |
| 376 |
specify this combination of flags. |
| 377 |
.P |
| 378 |
When a new namespace |
| 379 |
(other than a user namespace) |
| 380 |
is created via |
| 381 |
.BR clone (2) |
| 382 |
or |
| 383 |
.BR unshare (2), |
| 384 |
the kernel records the user namespace of the creating process |
| 385 |
as the owner of the new namespace. |
| 386 |
(This association can't be changed.) |
| 387 |
When a process in the new namespace subsequently performs |
| 388 |
privileged operations that operate |
| 389 |
on global resources isolated by the namespace, |
| 390 |
the permission checks are performed according to the process's capabilities |
| 391 |
in the user namespace that the kernel associated with the new namespace. |
| 392 |
For example, |
| 393 |
suppose that a process attempts to change the hostname |
| 394 |
.RB ( sethostname (2)), |
| 395 |
a resource governed by the UTS namespace. |
| 396 |
In this case, |
| 397 |
the kernel will determine |
| 398 |
which user namespace owns |
| 399 |
the process's UTS namespace, |
| 400 |
and check |
| 401 |
whether the process has the required capability |
| 402 |
.RB ( CAP_SYS_ADMIN ) |
| 403 |
in that user namespace. |
| 404 |
.P |
| 405 |
The |
| 406 |
.B NS_GET_USERNS |
| 407 |
.BR ioctl (2) |
| 408 |
operation can be used to discover |
| 409 |
the user namespace that owns a nonuser namespace; |
| 410 |
see |
| 411 |
.BR ioctl_nsfs (2). |
| 412 |
.\" |
| 413 |
.\" ============================================================ |
| 414 |
.\" |
| 415 |
.SS User and group ID mappings: uid_map and gid_map |
| 416 |
When a user namespace is created, |
| 417 |
it starts out without a mapping of user IDs (group IDs) |
| 418 |
to the parent user namespace. |
| 419 |
The |
| 420 |
.IR /proc/ pid /uid_map |
| 421 |
and |
| 422 |
.IR /proc/ pid /gid_map |
| 423 |
files (available since Linux 3.5) |
| 424 |
.\" commit 22d917d80e842829d0ca0a561967d728eb1d6303 |
| 425 |
expose the mappings for user and group IDs |
| 426 |
inside the user namespace |
| 427 |
for the process |
| 428 |
.IR pid . |
| 429 |
These files can be read to view the mappings in a user namespace |
| 430 |
and written to (once) to define the mappings. |
| 431 |
.P |
| 432 |
The description in the following paragraphs explains the details for |
| 433 |
.IR uid_map ; |
| 434 |
.I gid_map |
| 435 |
is exactly the same, |
| 436 |
but each instance of "user ID" is replaced by "group ID". |
| 437 |
.P |
| 438 |
The |
| 439 |
.I uid_map |
| 440 |
file exposes the mapping of user IDs |
| 441 |
from the user namespace of the process |
| 442 |
.I pid |
| 443 |
to the user namespace of the process that opened |
| 444 |
.I uid_map |
| 445 |
(but see a qualification to this point below). |
| 446 |
In other words, |
| 447 |
processes that are in different user namespaces |
| 448 |
will potentially see different values when reading from a particular |
| 449 |
.I uid_map |
| 450 |
file, |
| 451 |
depending on the user ID mappings |
| 452 |
for the user namespaces |
| 453 |
of the reading processes. |
| 454 |
.P |
| 455 |
Each line in the |
| 456 |
.I uid_map |
| 457 |
file specifies a 1-to-1 mapping of |
| 458 |
a range of contiguous user IDs between two user namespaces. |
| 459 |
(When a user namespace is first created, this file is empty.) |
| 460 |
The specification in each line takes the form of |
| 461 |
three numbers delimited by white space. |
| 462 |
The first two numbers specify |
| 463 |
the starting user ID |
| 464 |
in each of the two user namespaces. |
| 465 |
The third number specifies the size of the mapped range. |
| 466 |
In detail, |
| 467 |
the fields are interpreted as follows: |
| 468 |
.IP (1) 5 |
| 469 |
The start of the range of user IDs |
| 470 |
in the user namespace of the process |
| 471 |
.IR pid . |
| 472 |
.IP (2) |
| 473 |
The start of the range of user IDs |
| 474 |
to which the user IDs specified by field one map. |
| 475 |
How field two is interpreted depends on |
| 476 |
whether the process that opened |
| 477 |
.I uid_map |
| 478 |
and the process |
| 479 |
.I pid |
| 480 |
are in the same user namespace, |
| 481 |
as follows: |
| 482 |
.RS |
| 483 |
.IP (a) 5 |
| 484 |
If the two processes are in different user namespaces: |
| 485 |
field two is the start of a range of user IDs |
| 486 |
in the user namespace of the process that opened |
| 487 |
.IR uid_map . |
| 488 |
.IP (b) |
| 489 |
If the two processes are in the same user namespace: |
| 490 |
field two is the start of the range of user IDs |
| 491 |
in the parent user namespace of the process |
| 492 |
.IR pid . |
| 493 |
This case enables the opener of |
| 494 |
.I uid_map |
| 495 |
(the common case here is opening |
| 496 |
.IR /proc/self/uid_map ) |
| 497 |
to see the mapping of user IDs |
| 498 |
into the user namespace of the process that created this user namespace. |
| 499 |
.RE |
| 500 |
.IP (3) |
| 501 |
The size of the range of user IDs that is mapped between the two |
| 502 |
user namespaces. |
| 503 |
.P |
| 504 |
System calls that return user IDs (group IDs)\[em]for example, |
| 505 |
.BR getuid (2), |
| 506 |
.BR getgid (2), |
| 507 |
and the credential fields in the structure returned by |
| 508 |
.BR stat (2)\[em]return |
| 509 |
the user ID (group ID) mapped into the caller's user namespace. |
| 510 |
.P |
| 511 |
When a process accesses a file, |
| 512 |
its user and group IDs |
| 513 |
are mapped into the initial user namespace |
| 514 |
for the purpose of permission checking |
| 515 |
and assigning IDs when creating a file. |
| 516 |
When a process retrieves file user and group IDs via |
| 517 |
.BR stat (2), |
| 518 |
the IDs are mapped in the opposite direction, |
| 519 |
to produce values relative to the process user and group ID mappings. |
| 520 |
.P |
| 521 |
The initial user namespace has no parent namespace, |
| 522 |
but, for consistency, |
| 523 |
the kernel provides dummy user and group ID mapping files |
| 524 |
for this namespace. |
| 525 |
Looking at the |
| 526 |
.I uid_map |
| 527 |
file |
| 528 |
.RI ( gid_map |
| 529 |
is the same) |
| 530 |
from a shell in the initial namespace shows: |
| 531 |
.P |
| 532 |
.in +4n |
| 533 |
.EX |
| 534 |
.RB $ " cat /proc/$$/uid_map" ; |
| 535 |
0 0 4294967295 |
| 536 |
.EE |
| 537 |
.in |
| 538 |
.P |
| 539 |
This mapping tells us |
| 540 |
that the range starting at user ID 0 in this namespace |
| 541 |
maps to a range starting at 0 in the (nonexistent) parent namespace, |
| 542 |
and the size of the range is the largest 32-bit unsigned integer. |
| 543 |
This leaves 4294967295 (the 32-bit signed \-1 value) unmapped. |
| 544 |
This is deliberate: |
| 545 |
.I (uid_t)\~\-1 |
| 546 |
is used in several interfaces (e.g., |
| 547 |
.BR setreuid (2)) |
| 548 |
as a way to specify "no user ID". |
| 549 |
Leaving |
| 550 |
.I (uid_t)\~\-1 |
| 551 |
unmapped and unusable |
| 552 |
guarantees that there will be |
| 553 |
no confusion when using these interfaces. |
| 554 |
.\" |
| 555 |
.\" ============================================================ |
| 556 |
.\" |
| 557 |
.SS Defining user and group ID mappings: writing to uid_map and gid_map |
| 558 |
After the creation of a new user namespace, |
| 559 |
the |
| 560 |
.I uid_map |
| 561 |
file of |
| 562 |
.I one |
| 563 |
of the processes in the namespace may be written to |
| 564 |
.I once |
| 565 |
to define the mapping of user IDs in the new user namespace. |
| 566 |
An attempt to write more than once to a |
| 567 |
.I uid_map |
| 568 |
file in a user namespace fails with the error |
| 569 |
.BR EPERM . |
| 570 |
Similar rules apply for |
| 571 |
.I gid_map |
| 572 |
files. |
| 573 |
.P |
| 574 |
The lines written to |
| 575 |
.I uid_map |
| 576 |
.RI ( gid_map ) |
| 577 |
must conform to the following validity rules: |
| 578 |
.IP \[bu] 3 |
| 579 |
The three fields must be valid numbers, |
| 580 |
and the last field must be greater than 0. |
| 581 |
.IP \[bu] |
| 582 |
Lines are terminated by newline characters. |
| 583 |
.IP \[bu] |
| 584 |
There is a limit on the number of lines in the file. |
| 585 |
Up to Linux 4.14, |
| 586 |
.\" 5*12-byte records could fit in a 64B cache line |
| 587 |
this limit was (arbitrarily) set at 5 lines. |
| 588 |
Since Linux 4.16, |
| 589 |
.\" commit 6397fac4915ab3002dc15aae751455da1a852f25 |
| 590 |
the limit is 340 lines. |
| 591 |
In addition, |
| 592 |
the number of bytes written to the file |
| 593 |
must be less than the system page size, |
| 594 |
and the write must be performed at the start of the file (i.e., |
| 595 |
.BR lseek (2) |
| 596 |
and |
| 597 |
.BR pwrite (2) |
| 598 |
can't be used to write to nonzero offsets in the file). |
| 599 |
.IP \[bu] |
| 600 |
The range of user IDs (group IDs) |
| 601 |
specified in each line cannot overlap with the ranges |
| 602 |
in any other lines. |
| 603 |
In the initial implementation (Linux 3.8), |
| 604 |
this requirement was satisfied by a simplistic implementation |
| 605 |
that imposed the further requirement that |
| 606 |
the values in both field 1 and field 2 of successive lines |
| 607 |
must be in ascending numerical order, |
| 608 |
which prevented some otherwise valid maps from being created. |
| 609 |
Since Linux 3.9, |
| 610 |
.\" commit 0bd14b4fd72afd5df41e9fd59f356740f22fceba |
| 611 |
this limitation has been removed, |
| 612 |
allowing any valid set of nonoverlapping maps. |
| 613 |
.IP \[bu] |
| 614 |
At least one line must be written to the file. |
| 615 |
.P |
| 616 |
Writes that violate the above rules fail with the error |
| 617 |
.BR EINVAL . |
| 618 |
.P |
| 619 |
In order for a process to write to the |
| 620 |
.IR /proc/ pid /uid_map |
| 621 |
.RI ( /proc/ pid /gid_map ) |
| 622 |
file, |
| 623 |
all of the following permission requirements must be met: |
| 624 |
.IP \[bu] 3 |
| 625 |
The writing process must have the |
| 626 |
.B CAP_SETUID |
| 627 |
.RB ( CAP_SETGID ) |
| 628 |
capability in the user namespace of the process |
| 629 |
.IR pid . |
| 630 |
.IP \[bu] |
| 631 |
The writing process must either be in the user namespace of the process |
| 632 |
.I pid |
| 633 |
or be in the parent user namespace of the process |
| 634 |
.IR pid . |
| 635 |
.IP \[bu] |
| 636 |
The mapped user IDs (group IDs) must in turn |
| 637 |
have a mapping |
| 638 |
in the parent user namespace. |
| 639 |
.IP \[bu] |
| 640 |
If updating |
| 641 |
.IR /proc/ pid /uid_map |
| 642 |
to create a mapping that maps UID 0 in the parent namespace, |
| 643 |
then one of the following must be true: |
| 644 |
.RS |
| 645 |
.IP (a) 5 |
| 646 |
if writing process is in the parent user namespace, |
| 647 |
then it must have the |
| 648 |
.B CAP_SETFCAP |
| 649 |
capability in that user namespace; |
| 650 |
or |
| 651 |
.IP (b) |
| 652 |
if the writing process is in the child user namespace, |
| 653 |
then the process that created the user namespace must have had the |
| 654 |
.B CAP_SETFCAP |
| 655 |
capability when the namespace was created. |
| 656 |
.RE |
| 657 |
.IP |
| 658 |
This rule has been in place since |
| 659 |
.\" commit db2e718a47984b9d71ed890eb2ea36ecf150de18 |
| 660 |
Linux 5.12. |
| 661 |
It eliminates an earlier security bug whereby |
| 662 |
a UID 0 process that lacks the |
| 663 |
.B CAP_SETFCAP |
| 664 |
capability, |
| 665 |
which is needed to create a binary with namespaced file capabilities |
| 666 |
(as described in |
| 667 |
.BR capabilities (7)), |
| 668 |
could nevertheless create such a binary, |
| 669 |
by the following steps: |
| 670 |
.RS |
| 671 |
.IP (1) 5 |
| 672 |
Create a new user namespace with the identity mapping |
| 673 |
(i.e., UID 0 in the new user namespace maps to UID 0 in the parent namespace), |
| 674 |
so that UID 0 in both namespaces is equivalent to the same root user ID. |
| 675 |
.IP (2) |
| 676 |
Since the child process has the |
| 677 |
.B CAP_SETFCAP |
| 678 |
capability, |
| 679 |
it could create a binary with namespaced file capabilities |
| 680 |
that would then be effective in the parent user namespace |
| 681 |
(because the root user IDs are the same in the two namespaces). |
| 682 |
.RE |
| 683 |
.IP \[bu] |
| 684 |
One of the following two cases applies: |
| 685 |
.RS |
| 686 |
.IP (a) 5 |
| 687 |
.I Either |
| 688 |
the writing process has the |
| 689 |
.B CAP_SETUID |
| 690 |
.RB ( CAP_SETGID ) |
| 691 |
capability in the |
| 692 |
.I parent |
| 693 |
user namespace. |
| 694 |
.RS |
| 695 |
.IP \[bu] 3 |
| 696 |
No further restrictions apply: |
| 697 |
the process can make mappings to arbitrary user IDs (group IDs) |
| 698 |
in the parent user namespace. |
| 699 |
.RE |
| 700 |
.IP (b) |
| 701 |
.I Or |
| 702 |
otherwise all of the following restrictions apply: |
| 703 |
.RS |
| 704 |
.IP \[bu] 3 |
| 705 |
The data written to |
| 706 |
.I uid_map |
| 707 |
.RI ( gid_map ) |
| 708 |
must consist of a single line that maps |
| 709 |
the writing process's effective user ID (group ID) |
| 710 |
in the parent user namespace |
| 711 |
to a user ID (group ID) |
| 712 |
in the user namespace. |
| 713 |
.IP \[bu] |
| 714 |
The writing process must have the same effective user ID |
| 715 |
as the process that created the user namespace. |
| 716 |
.IP \[bu] |
| 717 |
In the case of |
| 718 |
.IR gid_map , |
| 719 |
use of the |
| 720 |
.BR setgroups (2) |
| 721 |
system call must first be denied by writing |
| 722 |
.RI \[dq] deny \[dq] |
| 723 |
to the |
| 724 |
.IR /proc/ pid /setgroups |
| 725 |
file (see below) |
| 726 |
before writing to |
| 727 |
.IR gid_map . |
| 728 |
.RE |
| 729 |
.RE |
| 730 |
.P |
| 731 |
Writes that violate the above rules fail with the error |
| 732 |
.BR EPERM . |
| 733 |
.\" |
| 734 |
.\" ============================================================ |
| 735 |
.\" |
| 736 |
.SS Project ID mappings: projid_map |
| 737 |
Similarly to user and group ID mappings, |
| 738 |
it is possible to create project ID mappings for a user namespace. |
| 739 |
(Project IDs are used for disk quotas; |
| 740 |
see |
| 741 |
.BR setquota (8) |
| 742 |
and |
| 743 |
.BR quotactl (2).) |
| 744 |
.P |
| 745 |
Project ID mappings are defined by writing to the |
| 746 |
.IR /proc/ pid /projid_map |
| 747 |
file |
| 748 |
.\" commit f76d207a66c3a53defea67e7d36c3eb1b7d6d61d |
| 749 |
(present since Linux 3.7). |
| 750 |
.P |
| 751 |
The validity rules for writing to the |
| 752 |
.IR /proc/ pid /projid_map |
| 753 |
file are as for writing to the |
| 754 |
.I uid_map |
| 755 |
file; |
| 756 |
violation of these rules causes |
| 757 |
.BR write (2) |
| 758 |
to fail with the error |
| 759 |
.BR EINVAL . |
| 760 |
.P |
| 761 |
The permission rules for writing to the |
| 762 |
.IR /proc/ pid /projid_map |
| 763 |
file are as follows: |
| 764 |
.IP \[bu] 3 |
| 765 |
The writing process must either be in the user namespace of the process |
| 766 |
.I pid |
| 767 |
or be in the parent user namespace of the process |
| 768 |
.IR pid . |
| 769 |
.IP \[bu] |
| 770 |
The mapped project IDs must in turn |
| 771 |
have a mapping |
| 772 |
in the parent user namespace. |
| 773 |
.P |
| 774 |
Violation of these rules causes |
| 775 |
.BR write (2) |
| 776 |
to fail with the error |
| 777 |
.BR EPERM . |
| 778 |
.\" |
| 779 |
.\" ============================================================ |
| 780 |
.\" |
| 781 |
.SS Interaction with system calls that change process UIDs or GIDs |
| 782 |
In a user namespace where the |
| 783 |
.I uid_map |
| 784 |
file has not been written, |
| 785 |
the system calls that change user IDs will fail. |
| 786 |
Similarly, |
| 787 |
if the |
| 788 |
.I gid_map |
| 789 |
file has not been written, |
| 790 |
the system calls that change group IDs will fail. |
| 791 |
After the |
| 792 |
.I uid_map |
| 793 |
and |
| 794 |
.I gid_map |
| 795 |
files have been written, |
| 796 |
only the mapped values may be used |
| 797 |
in system calls that change user and group IDs. |
| 798 |
.P |
| 799 |
For user IDs, |
| 800 |
the relevant system calls include |
| 801 |
.BR setuid (2), |
| 802 |
.BR setfsuid (2), |
| 803 |
.BR setreuid (2), |
| 804 |
and |
| 805 |
.BR setresuid (2). |
| 806 |
For group IDs, |
| 807 |
the relevant system calls include |
| 808 |
.BR setgid (2), |
| 809 |
.BR setfsgid (2), |
| 810 |
.BR setregid (2), |
| 811 |
.BR setresgid (2), |
| 812 |
and |
| 813 |
.BR setgroups (2). |
| 814 |
.P |
| 815 |
Writing |
| 816 |
.RI \[dq] deny \[dq] |
| 817 |
to the |
| 818 |
.IR /proc/ pid /setgroups |
| 819 |
file before writing to |
| 820 |
.IR /proc/ pid /gid_map |
| 821 |
.\" Things changed in Linux 3.19 |
| 822 |
.\" commit 9cc46516ddf497ea16e8d7cb986ae03a0f6b92f8 |
| 823 |
.\" commit 66d2f338ee4c449396b6f99f5e75cd18eb6df272 |
| 824 |
.\" http://lwn.net/Articles/626665/ |
| 825 |
will permanently disable |
| 826 |
.BR setgroups (2) |
| 827 |
in a user namespace and allow writing to |
| 828 |
.IR /proc/ pid /gid_map |
| 829 |
without having the |
| 830 |
.B CAP_SETGID |
| 831 |
capability in the parent user namespace. |
| 832 |
.\" |
| 833 |
.\" ============================================================ |
| 834 |
.\" |
| 835 |
.SS The \f[I]/proc/\f[]pid\f[I]/setgroups\f[] file |
| 836 |
.\" |
| 837 |
.\" commit 9cc46516ddf497ea16e8d7cb986ae03a0f6b92f8 |
| 838 |
.\" commit 66d2f338ee4c449396b6f99f5e75cd18eb6df272 |
| 839 |
.\" http://lwn.net/Articles/626665/ |
| 840 |
.\" http://web.nvd.nist.gov/view/vuln/detail?vulnId=CVE-2014-8989 |
| 841 |
.\" |
| 842 |
The |
| 843 |
.IR /proc/ pid /setgroups |
| 844 |
file displays the string |
| 845 |
.RI \[dq] allow \[dq] |
| 846 |
if processes in the user namespace that contains the process |
| 847 |
.I pid |
| 848 |
are permitted to employ the |
| 849 |
.BR setgroups (2) |
| 850 |
system call; |
| 851 |
it displays |
| 852 |
.RI \[dq] deny \[dq] |
| 853 |
if |
| 854 |
.BR setgroups (2) |
| 855 |
is not permitted in that user namespace. |
| 856 |
Note that regardless of the value in the |
| 857 |
.IR /proc/ pid /setgroups |
| 858 |
file (and regardless of the process's capabilities), |
| 859 |
calls to |
| 860 |
.BR setgroups (2) |
| 861 |
are also not permitted if |
| 862 |
.IR /proc/ pid /gid_map |
| 863 |
has not yet been set. |
| 864 |
.P |
| 865 |
A privileged process |
| 866 |
(one with the |
| 867 |
.B CAP_SYS_ADMIN |
| 868 |
capability in the namespace) |
| 869 |
may write either of the strings |
| 870 |
.RI \[dq] allow \[dq] |
| 871 |
or |
| 872 |
.RI \[dq] deny \[dq] |
| 873 |
to this file |
| 874 |
.I before |
| 875 |
writing a group ID mapping |
| 876 |
for this user namespace to the file |
| 877 |
.IR /proc/ pid /gid_map . |
| 878 |
Writing the string |
| 879 |
.RI \[dq] deny \[dq] |
| 880 |
prevents any process in the user namespace from employing |
| 881 |
.BR setgroups (2). |
| 882 |
.P |
| 883 |
The essence of the restrictions described in the preceding paragraph is that |
| 884 |
it is permitted to write to |
| 885 |
.IR /proc/ pid /setgroups |
| 886 |
only so long as calling |
| 887 |
.BR setgroups (2) |
| 888 |
is disallowed because |
| 889 |
.IR /proc/ pid /gid_map |
| 890 |
has not been set. |
| 891 |
This ensures that a process cannot transition |
| 892 |
from a state where |
| 893 |
.BR setgroups (2) |
| 894 |
is allowed |
| 895 |
to a state where |
| 896 |
.BR setgroups (2) |
| 897 |
is denied; |
| 898 |
a process can transition only |
| 899 |
from |
| 900 |
.BR setgroups (2) |
| 901 |
being disallowed |
| 902 |
to |
| 903 |
.BR setgroups (2) |
| 904 |
being allowed. |
| 905 |
.P |
| 906 |
The default value of this file in the initial user namespace is |
| 907 |
.RI \[dq] allow \[dq]. |
| 908 |
.P |
| 909 |
Once |
| 910 |
.IR /proc/ pid /gid_map |
| 911 |
has been written to |
| 912 |
(which has the effect of enabling |
| 913 |
.BR setgroups (2) |
| 914 |
in the user namespace), |
| 915 |
it is no longer possible to disallow |
| 916 |
.BR setgroups (2) |
| 917 |
by writing |
| 918 |
.RI \[dq] deny \[dq] |
| 919 |
to |
| 920 |
.IR /proc/ pid /setgroups |
| 921 |
(the write fails with the error |
| 922 |
.BR EPERM ). |
| 923 |
.P |
| 924 |
A child user namespace inherits the |
| 925 |
.IR /proc/ pid /setgroups |
| 926 |
setting from its parent. |
| 927 |
.P |
| 928 |
If the |
| 929 |
.I setgroups |
| 930 |
file has the value |
| 931 |
.RI \[dq] deny \[dq], |
| 932 |
then the |
| 933 |
.BR setgroups (2) |
| 934 |
system call can't subsequently be reenabled |
| 935 |
(by writing |
| 936 |
.RI \[dq] allow \[dq] |
| 937 |
to the file) |
| 938 |
in this user namespace. |
| 939 |
(Attempts to do so fail with the error |
| 940 |
.BR EPERM .) |
| 941 |
This restriction also propagates down |
| 942 |
to all child user namespaces of this user namespace. |
| 943 |
.P |
| 944 |
The |
| 945 |
.IR /proc/ pid /setgroups |
| 946 |
file was added in Linux 3.19, |
| 947 |
but was backported to many earlier stable kernel series, |
| 948 |
because it addresses a security issue. |
| 949 |
The issue concerned files with permissions such as "rwx\-\-\-rwx". |
| 950 |
Such files give fewer permissions to "group" than they do to "other". |
| 951 |
This means that dropping groups using |
| 952 |
.BR setgroups (2) |
| 953 |
might allow a process file access that it did not formerly have. |
| 954 |
Before the existence of user namespaces this was not a concern, |
| 955 |
since only a privileged process |
| 956 |
(one with the |
| 957 |
.B CAP_SETGID |
| 958 |
capability) |
| 959 |
could call |
| 960 |
.BR setgroups (2). |
| 961 |
However, |
| 962 |
with the introduction of user namespaces, |
| 963 |
it became possible for an unprivileged process to |
| 964 |
create a new namespace in which the user had all privileges. |
| 965 |
This then allowed formerly unprivileged |
| 966 |
users to drop groups and thus gain file access |
| 967 |
that they did not previously have. |
| 968 |
The |
| 969 |
.IR /proc/ pid /setgroups |
| 970 |
file was added to address this security issue, |
| 971 |
by denying any pathway for an unprivileged process to drop groups with |
| 972 |
.BR setgroups (2). |
| 973 |
.\" |
| 974 |
.\" /proc/PID/setgroups |
| 975 |
.\" [allow == setgroups() is allowed, "deny" == setgroups() is disallowed] |
| 976 |
.\" * Can write if have CAP_SYS_ADMIN in NS |
| 977 |
.\" * Must write BEFORE writing to /proc/PID/gid_map |
| 978 |
.\" |
| 979 |
.\" setgroups() |
| 980 |
.\" * Must already have written to gid_map |
| 981 |
.\" * /proc/PID/setgroups must be "allow" |
| 982 |
.\" |
| 983 |
.\" /proc/PID/gid_map -- writing |
| 984 |
.\" * Must already have written "deny" to /proc/PID/setgroups |
| 985 |
.\" |
| 986 |
.\" ============================================================ |
| 987 |
.\" |
| 988 |
.SS Unmapped user and group IDs |
| 989 |
There are various places where an unmapped user ID (group ID) |
| 990 |
may be exposed to user space. |
| 991 |
For example, |
| 992 |
the first process in a new user namespace may call |
| 993 |
.BR getuid (2) |
| 994 |
before a user ID mapping has been defined for the namespace. |
| 995 |
In most such cases, |
| 996 |
an unmapped user ID is converted |
| 997 |
.\" from_kuid_munged(), from_kgid_munged() |
| 998 |
to the overflow user ID (group ID); |
| 999 |
the default value for the overflow user ID (group ID) is 65534. |
| 1000 |
See the descriptions of |
| 1001 |
.I /proc/sys/kernel/overflowuid |
| 1002 |
and |
| 1003 |
.I /proc/sys/kernel/overflowgid |
| 1004 |
in |
| 1005 |
.BR proc (5). |
| 1006 |
.P |
| 1007 |
The cases where unmapped IDs are mapped in this fashion include |
| 1008 |
system calls that return user IDs |
| 1009 |
.RB ( getuid (2), |
| 1010 |
.BR getgid (2), |
| 1011 |
and similar), |
| 1012 |
credentials passed over a UNIX domain socket, |
| 1013 |
.\" also SO_PEERCRED |
| 1014 |
credentials returned by |
| 1015 |
.BR stat (2), |
| 1016 |
.BR waitid (2), |
| 1017 |
and the System V IPC "ctl" |
| 1018 |
.B IPC_STAT |
| 1019 |
operations, |
| 1020 |
credentials exposed by |
| 1021 |
.IR /proc/ pid /status |
| 1022 |
and the files in |
| 1023 |
.IR /proc/sysvipc/* , |
| 1024 |
credentials returned via the |
| 1025 |
.I si_uid |
| 1026 |
field in the |
| 1027 |
.I siginfo_t |
| 1028 |
received with a signal |
| 1029 |
(see |
| 1030 |
.BR sigaction (2)), |
| 1031 |
credentials written to the process accounting file |
| 1032 |
(see |
| 1033 |
.BR acct (5)), |
| 1034 |
and credentials returned with POSIX message queue notifications |
| 1035 |
(see |
| 1036 |
.BR mq_notify (3)). |
| 1037 |
.P |
| 1038 |
There is one notable case where unmapped user and group IDs are |
| 1039 |
.I not |
| 1040 |
.\" from_kuid(), from_kgid() |
| 1041 |
.\" Also F_GETOWNER_UIDS is an exception |
| 1042 |
converted to the corresponding overflow ID value. |
| 1043 |
When viewing a |
| 1044 |
.I uid_map |
| 1045 |
or |
| 1046 |
.I gid_map |
| 1047 |
file in which there is no mapping for the second field, |
| 1048 |
that field is displayed as 4294967295 |
| 1049 |
(\-1 as an unsigned integer). |
| 1050 |
.\" |
| 1051 |
.\" ============================================================ |
| 1052 |
.\" |
| 1053 |
.SS Accessing files |
| 1054 |
In order to determine permissions when an unprivileged process accesses a file, |
| 1055 |
the process credentials (UID, GID) and the file credentials |
| 1056 |
are in effect mapped back to |
| 1057 |
what they would be in the initial user namespace |
| 1058 |
and then compared to determine |
| 1059 |
the permissions that the process has on the file. |
| 1060 |
The same is also true of other objects that |
| 1061 |
employ the credentials plus permissions mask accessibility model, |
| 1062 |
such as System V IPC objects. |
| 1063 |
.\" |
| 1064 |
.\" ============================================================ |
| 1065 |
.\" |
| 1066 |
.SS Operation of file-related capabilities |
| 1067 |
Certain capabilities allow a process to |
| 1068 |
bypass various kernel-enforced restrictions |
| 1069 |
when performing operations on |
| 1070 |
files owned by other users or groups. |
| 1071 |
These capabilities are: |
| 1072 |
.BR CAP_CHOWN , |
| 1073 |
.BR CAP_DAC_OVERRIDE , |
| 1074 |
.BR CAP_DAC_READ_SEARCH , |
| 1075 |
.BR CAP_FOWNER , |
| 1076 |
and |
| 1077 |
.BR CAP_FSETID . |
| 1078 |
.P |
| 1079 |
Within a user namespace, |
| 1080 |
these capabilities allow a process to bypass the rules |
| 1081 |
if the process has the relevant capability over the file, |
| 1082 |
meaning that: |
| 1083 |
.IP \[bu] 3 |
| 1084 |
the process has the relevant effective capability in its user namespace; |
| 1085 |
and |
| 1086 |
.IP \[bu] |
| 1087 |
the file's user ID and group ID both have valid mappings |
| 1088 |
in the user namespace. |
| 1089 |
.P |
| 1090 |
The |
| 1091 |
.B CAP_FOWNER |
| 1092 |
capability is treated somewhat exceptionally: |
| 1093 |
.\" These are the checks performed by the kernel function |
| 1094 |
.\" inode_owner_or_capable(). There is one exception to the exception: |
| 1095 |
.\" overriding the directory sticky permission bit requires that |
| 1096 |
.\" the file has a valid mapping for both its UID and GID. |
| 1097 |
it allows a process to bypass the corresponding rules so long as |
| 1098 |
at least the file's user ID has a mapping in the user namespace |
| 1099 |
(i.e., the file's group ID does not need to have a valid mapping). |
| 1100 |
.\" |
| 1101 |
.\" ============================================================ |
| 1102 |
.\" |
| 1103 |
.SS Set-user-ID and set-group-ID programs |
| 1104 |
When a process inside a user namespace |
| 1105 |
executes a set-user-ID (set-group-ID) program, |
| 1106 |
the process's effective user (group) ID inside the namespace |
| 1107 |
is changed to whatever value is mapped for the user (group) ID of the file. |
| 1108 |
However, |
| 1109 |
if either the user |
| 1110 |
.I or |
| 1111 |
the group ID of the file has no mapping inside the namespace, |
| 1112 |
the set-user-ID (set-group-ID) bit is silently ignored: |
| 1113 |
the new program is executed, |
| 1114 |
but the process's effective user (group) ID is left unchanged. |
| 1115 |
(This mirrors the semantics of |
| 1116 |
executing a set-user-ID or set-group-ID program |
| 1117 |
that resides on a filesystem that was mounted with the |
| 1118 |
.B MS_NOSUID |
| 1119 |
flag, as described in |
| 1120 |
.BR mount (2).) |
| 1121 |
.\" |
| 1122 |
.\" ============================================================ |
| 1123 |
.\" |
| 1124 |
.SS Miscellaneous |
| 1125 |
When a process's user and group IDs are passed over a UNIX domain socket |
| 1126 |
to a process in a different user namespace |
| 1127 |
(see the description of |
| 1128 |
.B SCM_CREDENTIALS |
| 1129 |
in |
| 1130 |
.BR unix (7)), |
| 1131 |
they are translated into the corresponding values |
| 1132 |
as per the receiving process's user and group ID mappings. |
| 1133 |
.\" |
| 1134 |
.SH STANDARDS |
| 1135 |
Linux. |
| 1136 |
.\" |
| 1137 |
.SH NOTES |
| 1138 |
Over the years, |
| 1139 |
there have been a lot of |
| 1140 |
features that have been added to the Linux kernel |
| 1141 |
that have been made available only to privileged users |
| 1142 |
because of their potential to confuse set-user-ID-root applications. |
| 1143 |
In general, |
| 1144 |
it becomes safe to |
| 1145 |
allow the root user in a user namespace to use those features |
| 1146 |
because it is impossible, |
| 1147 |
while in a user namespace, |
| 1148 |
to gain more privilege than the root user of a user namespace has. |
| 1149 |
.\" |
| 1150 |
.\" ============================================================ |
| 1151 |
.\" |
| 1152 |
.SS Global root |
| 1153 |
The term "global root" is sometimes used as a shorthand for |
| 1154 |
user ID 0 in the initial user namespace. |
| 1155 |
.\" |
| 1156 |
.\" ============================================================ |
| 1157 |
.\" |
| 1158 |
.SS Availability |
| 1159 |
Use of user namespaces requires a kernel that is configured with the |
| 1160 |
.B CONFIG_USER_NS |
| 1161 |
option. |
| 1162 |
User namespaces require support |
| 1163 |
in a range of subsystems across the kernel. |
| 1164 |
When an unsupported subsystem is configured into the kernel, |
| 1165 |
it is not possible to configure user namespaces support. |
| 1166 |
.P |
| 1167 |
As at Linux 3.8, |
| 1168 |
most relevant subsystems supported user namespaces, |
| 1169 |
but a number of filesystems did not have the infrastructure |
| 1170 |
needed to map user and group IDs between user namespaces. |
| 1171 |
Linux 3.9 added the required infrastructure support |
| 1172 |
for many of the remaining unsupported filesystems |
| 1173 |
(Plan 9 (9P), Andrew File System (AFS), Ceph, CIFS, CODA, NFS, and OCFS2). |
| 1174 |
Linux 3.12 added support for the last of the unsupported major filesystems, |
| 1175 |
.\" commit d6970d4b726cea6d7a9bc4120814f95c09571fc3 |
| 1176 |
XFS. |
| 1177 |
.\" |
| 1178 |
.SH EXAMPLES |
| 1179 |
The program below is designed to allow experimenting |
| 1180 |
with user namespaces, |
| 1181 |
as well as other types of namespaces. |
| 1182 |
It creates namespaces as specified by command-line options |
| 1183 |
and then executes a command inside those namespaces. |
| 1184 |
The comments and |
| 1185 |
.IR usage () |
| 1186 |
function inside the program provide a full explanation of the program. |
| 1187 |
The following shell session demonstrates its use. |
| 1188 |
.P |
| 1189 |
First, we look at the run-time environment: |
| 1190 |
.P |
| 1191 |
.in +4n |
| 1192 |
.EX |
| 1193 |
.RB $ " uname \-rs" "; # Need Linux 3.8 or later" |
| 1194 |
Linux 3.8.0 |
| 1195 |
.RB $ " id \-u" "; # Running as unprivileged user" |
| 1196 |
1000 |
| 1197 |
.RB $ " id \-g" ; |
| 1198 |
1000 |
| 1199 |
.EE |
| 1200 |
.in |
| 1201 |
.P |
| 1202 |
Now start a new shell in new user |
| 1203 |
.RI ( \-U ), |
| 1204 |
mount |
| 1205 |
.RI ( \-m ), |
| 1206 |
and PID |
| 1207 |
.RI ( \-p ) |
| 1208 |
namespaces, with user ID |
| 1209 |
.RI ( \-M ) |
| 1210 |
and group ID |
| 1211 |
.RI ( \-G ) |
| 1212 |
1000 mapped to 0 inside the user namespace: |
| 1213 |
.P |
| 1214 |
.in +4n |
| 1215 |
.EX |
| 1216 |
.RB $ " ./userns_child_exec \-p \-m \-U \-M \[aq]0 1000 1\[aq] \-G \[aq]0 1000 1\[aq] bash" ; |
| 1217 |
.EE |
| 1218 |
.in |
| 1219 |
.P |
| 1220 |
The shell has PID 1, because it is the first process in the new |
| 1221 |
PID namespace: |
| 1222 |
.P |
| 1223 |
.in +4n |
| 1224 |
.EX |
| 1225 |
.RB bash$ " echo $$" ; |
| 1226 |
1 |
| 1227 |
.EE |
| 1228 |
.in |
| 1229 |
.P |
| 1230 |
Mounting a new |
| 1231 |
.I /proc |
| 1232 |
filesystem and listing all of the processes |
| 1233 |
visible in the new PID namespace |
| 1234 |
shows that |
| 1235 |
the shell can't see |
| 1236 |
any processes outside the PID namespace: |
| 1237 |
.P |
| 1238 |
.in +4n |
| 1239 |
.EX |
| 1240 |
.RB bash$ " mount \-t proc proc /proc" ; |
| 1241 |
.RB bash$ " ps ax" ; |
| 1242 |
PID TTY STAT TIME COMMAND |
| 1243 |
1 pts/3 S 0:00 bash |
| 1244 |
22 pts/3 R+ 0:00 ps ax |
| 1245 |
.EE |
| 1246 |
.in |
| 1247 |
.P |
| 1248 |
Inside the user namespace, |
| 1249 |
the shell has user and group ID 0, |
| 1250 |
and a full set of permitted and effective capabilities: |
| 1251 |
.P |
| 1252 |
.in +4n |
| 1253 |
.EX |
| 1254 |
.RB bash$ " cat /proc/$$/status | egrep \[aq]\[ha][UG]id\[aq]" ; |
| 1255 |
Uid: 0 0 0 0 |
| 1256 |
Gid: 0 0 0 0 |
| 1257 |
.RB bash$ " cat /proc/$$/status | egrep \[aq]\[ha]Cap(Prm|Inh|Eff)\[aq]" ; |
| 1258 |
CapInh: 0000000000000000 |
| 1259 |
CapPrm: 0000001fffffffff |
| 1260 |
CapEff: 0000001fffffffff |
| 1261 |
.EE |
| 1262 |
.in |
| 1263 |
.SS Program source |
| 1264 |
\& |
| 1265 |
.EX |
| 1266 |
/* userns_child_exec.c |
| 1267 |
\& |
| 1268 |
Licensed under GNU General Public License v2 or later |
| 1269 |
\& |
| 1270 |
Create a child process that executes a shell command in new |
| 1271 |
namespace(s); allow UID and GID mappings to be specified when |
| 1272 |
creating a user namespace. |
| 1273 |
*/ |
| 1274 |
#define _GNU_SOURCE |
| 1275 |
#include <err.h> |
| 1276 |
#include <sched.h> |
| 1277 |
#include <unistd.h> |
| 1278 |
#include <stdint.h> |
| 1279 |
#include <stdlib.h> |
| 1280 |
#include <sys/wait.h> |
| 1281 |
#include <signal.h> |
| 1282 |
#include <fcntl.h> |
| 1283 |
#include <stdio.h> |
| 1284 |
#include <string.h> |
| 1285 |
#include <limits.h> |
| 1286 |
#include <errno.h> |
| 1287 |
\& |
| 1288 |
struct child_args { |
| 1289 |
char **argv; /* Command to be executed by child, with args */ |
| 1290 |
int pipe_fd[2]; /* Pipe used to synchronize parent and child */ |
| 1291 |
}; |
| 1292 |
\& |
| 1293 |
static int verbose; |
| 1294 |
\& |
| 1295 |
static void |
| 1296 |
usage(char *pname) |
| 1297 |
{ |
| 1298 |
fprintf(stderr, "Usage: %s [options] cmd [arg...]\[rs]n\[rs]n", pname); |
| 1299 |
fprintf(stderr, "Create a child process that executes a shell " |
| 1300 |
"command in a new user namespace,\[rs]n" |
| 1301 |
"and possibly also other new namespace(s).\[rs]n\[rs]n"); |
| 1302 |
fprintf(stderr, "Options can be:\[rs]n\[rs]n"); |
| 1303 |
#define fpe(str) fprintf(stderr, " %s", str); |
| 1304 |
fpe("\-i New IPC namespace\[rs]n"); |
| 1305 |
fpe("\-m New mount namespace\[rs]n"); |
| 1306 |
fpe("\-n New network namespace\[rs]n"); |
| 1307 |
fpe("\-p New PID namespace\[rs]n"); |
| 1308 |
fpe("\-u New UTS namespace\[rs]n"); |
| 1309 |
fpe("\-U New user namespace\[rs]n"); |
| 1310 |
fpe("\-M uid_map Specify UID map for user namespace\[rs]n"); |
| 1311 |
fpe("\-G gid_map Specify GID map for user namespace\[rs]n"); |
| 1312 |
fpe("\-z Map user\[aq]s UID and GID to 0 in user namespace\[rs]n"); |
| 1313 |
fpe(" (equivalent to: \-M \[aq]0 <uid> 1\[aq] \-G \[aq]0 <gid> 1\[aq])\[rs]n"); |
| 1314 |
fpe("\-v Display verbose messages\[rs]n"); |
| 1315 |
fpe("\[rs]n"); |
| 1316 |
fpe("If \-z, \-M, or \-G is specified, \-U is required.\[rs]n"); |
| 1317 |
fpe("It is not permitted to specify both \-z and either \-M or \-G.\[rs]n"); |
| 1318 |
fpe("\[rs]n"); |
| 1319 |
fpe("Map strings for \-M and \-G consist of records of the form:\[rs]n"); |
| 1320 |
fpe("\[rs]n"); |
| 1321 |
fpe(" ID\-inside\-ns ID\-outside\-ns size\[rs]n"); |
| 1322 |
fpe("\[rs]n"); |
| 1323 |
fpe("A map string can contain multiple records, separated" |
| 1324 |
" by commas;\[rs]n"); |
| 1325 |
fpe("the commas are replaced by newlines before writing" |
| 1326 |
" to map files.\[rs]n"); |
| 1327 |
\& |
| 1328 |
exit(EXIT_FAILURE); |
| 1329 |
} |
| 1330 |
\& |
| 1331 |
/* Update the mapping file \[aq]map_file\[aq], with the value provided in |
| 1332 |
\[aq]mapping\[aq], a string that defines a UID or GID mapping. A UID or |
| 1333 |
GID mapping consists of one or more newline\-delimited records |
| 1334 |
of the form: |
| 1335 |
\& |
| 1336 |
ID_inside\-ns ID\-outside\-ns size |
| 1337 |
\& |
| 1338 |
Requiring the user to supply a string that contains newlines is |
| 1339 |
of course inconvenient for command\-line use. Thus, we permit the |
| 1340 |
use of commas to delimit records in this string, and replace them |
| 1341 |
with newlines before writing the string to the file. */ |
| 1342 |
\& |
| 1343 |
static void |
| 1344 |
update_map(char *mapping, char *map_file) |
| 1345 |
{ |
| 1346 |
int fd; |
| 1347 |
size_t map_len; /* Length of \[aq]mapping\[aq] */ |
| 1348 |
\& |
| 1349 |
/* Replace commas in mapping string with newlines. */ |
| 1350 |
\& |
| 1351 |
map_len = strlen(mapping); |
| 1352 |
for (size_t j = 0; j < map_len; j++) |
| 1353 |
if (mapping[j] == \[aq],\[aq]) |
| 1354 |
mapping[j] = \[aq]\[rs]n\[aq]; |
| 1355 |
\& |
| 1356 |
fd = open(map_file, O_RDWR); |
| 1357 |
if (fd == \-1) { |
| 1358 |
fprintf(stderr, "ERROR: open %s: %s\[rs]n", map_file, |
| 1359 |
strerror(errno)); |
| 1360 |
exit(EXIT_FAILURE); |
| 1361 |
} |
| 1362 |
\& |
| 1363 |
if (write(fd, mapping, map_len) != map_len) { |
| 1364 |
fprintf(stderr, "ERROR: write %s: %s\[rs]n", map_file, |
| 1365 |
strerror(errno)); |
| 1366 |
exit(EXIT_FAILURE); |
| 1367 |
} |
| 1368 |
\& |
| 1369 |
close(fd); |
| 1370 |
} |
| 1371 |
\& |
| 1372 |
/* Linux 3.19 made a change in the handling of setgroups(2) and |
| 1373 |
the \[aq]gid_map\[aq] file to address a security issue. The issue |
| 1374 |
allowed *unprivileged* users to employ user namespaces in |
| 1375 |
order to drop groups. The upshot of the 3.19 changes is that |
| 1376 |
in order to update the \[aq]gid_maps\[aq] file, use of the setgroups() |
| 1377 |
system call in this user namespace must first be disabled by |
| 1378 |
writing "deny" to one of the /proc/PID/setgroups files for |
| 1379 |
this namespace. That is the purpose of the following function. */ |
| 1380 |
\& |
| 1381 |
static void |
| 1382 |
proc_setgroups_write(pid_t child_pid, char *str) |
| 1383 |
{ |
| 1384 |
char setgroups_path[PATH_MAX]; |
| 1385 |
int fd; |
| 1386 |
\& |
| 1387 |
snprintf(setgroups_path, PATH_MAX, "/proc/%jd/setgroups", |
| 1388 |
(intmax_t) child_pid); |
| 1389 |
\& |
| 1390 |
fd = open(setgroups_path, O_RDWR); |
| 1391 |
if (fd == \-1) { |
| 1392 |
\& |
| 1393 |
/* We may be on a system that doesn\[aq]t support |
| 1394 |
/proc/PID/setgroups. In that case, the file won\[aq]t exist, |
| 1395 |
and the system won\[aq]t impose the restrictions that Linux 3.19 |
| 1396 |
added. That\[aq]s fine: we don\[aq]t need to do anything in order |
| 1397 |
to permit \[aq]gid_map\[aq] to be updated. |
| 1398 |
\& |
| 1399 |
However, if the error from open() was something other than |
| 1400 |
the ENOENT error that is expected for that case, let the |
| 1401 |
user know. */ |
| 1402 |
\& |
| 1403 |
if (errno != ENOENT) |
| 1404 |
fprintf(stderr, "ERROR: open %s: %s\[rs]n", setgroups_path, |
| 1405 |
strerror(errno)); |
| 1406 |
return; |
| 1407 |
} |
| 1408 |
\& |
| 1409 |
if (write(fd, str, strlen(str)) == \-1) |
| 1410 |
fprintf(stderr, "ERROR: write %s: %s\[rs]n", setgroups_path, |
| 1411 |
strerror(errno)); |
| 1412 |
\& |
| 1413 |
close(fd); |
| 1414 |
} |
| 1415 |
\& |
| 1416 |
static int /* Start function for cloned child */ |
| 1417 |
childFunc(void *arg) |
| 1418 |
{ |
| 1419 |
struct child_args *args = arg; |
| 1420 |
char ch; |
| 1421 |
\& |
| 1422 |
/* Wait until the parent has updated the UID and GID mappings. |
| 1423 |
See the comment in main(). We wait for end of file on a |
| 1424 |
pipe that will be closed by the parent process once it has |
| 1425 |
updated the mappings. */ |
| 1426 |
\& |
| 1427 |
close(args\->pipe_fd[1]); /* Close our descriptor for the write |
| 1428 |
end of the pipe so that we see EOF |
| 1429 |
when parent closes its descriptor. */ |
| 1430 |
if (read(args\->pipe_fd[0], &ch, 1) != 0) { |
| 1431 |
fprintf(stderr, |
| 1432 |
"Failure in child: read from pipe returned != 0\[rs]n"); |
| 1433 |
exit(EXIT_FAILURE); |
| 1434 |
} |
| 1435 |
\& |
| 1436 |
close(args\->pipe_fd[0]); |
| 1437 |
\& |
| 1438 |
/* Execute a shell command. */ |
| 1439 |
\& |
| 1440 |
printf("About to exec %s\[rs]n", args\->argv[0]); |
| 1441 |
execvp(args\->argv[0], args\->argv); |
| 1442 |
err(EXIT_FAILURE, "execvp"); |
| 1443 |
} |
| 1444 |
\& |
| 1445 |
#define STACK_SIZE (1024 * 1024) |
| 1446 |
\& |
| 1447 |
static char child_stack[STACK_SIZE]; /* Space for child\[aq]s stack */ |
| 1448 |
\& |
| 1449 |
int |
| 1450 |
main(int argc, char *argv[]) |
| 1451 |
{ |
| 1452 |
int flags, opt, map_zero; |
| 1453 |
pid_t child_pid; |
| 1454 |
struct child_args args; |
| 1455 |
char *uid_map, *gid_map; |
| 1456 |
const int MAP_BUF_SIZE = 100; |
| 1457 |
char map_buf[MAP_BUF_SIZE]; |
| 1458 |
char map_path[PATH_MAX]; |
| 1459 |
\& |
| 1460 |
/* Parse command\-line options. The initial \[aq]+\[aq] character in |
| 1461 |
the final getopt() argument prevents GNU\-style permutation |
| 1462 |
of command\-line options. That\[aq]s useful, since sometimes |
| 1463 |
the \[aq]command\[aq] to be executed by this program itself |
| 1464 |
has command\-line options. We don\[aq]t want getopt() to treat |
| 1465 |
those as options to this program. */ |
| 1466 |
\& |
| 1467 |
flags = 0; |
| 1468 |
verbose = 0; |
| 1469 |
gid_map = NULL; |
| 1470 |
uid_map = NULL; |
| 1471 |
map_zero = 0; |
| 1472 |
while ((opt = getopt(argc, argv, "+imnpuUM:G:zv")) != \-1) { |
| 1473 |
switch (opt) { |
| 1474 |
case \[aq]i\[aq]: flags |= CLONE_NEWIPC; break; |
| 1475 |
case \[aq]m\[aq]: flags |= CLONE_NEWNS; break; |
| 1476 |
case \[aq]n\[aq]: flags |= CLONE_NEWNET; break; |
| 1477 |
case \[aq]p\[aq]: flags |= CLONE_NEWPID; break; |
| 1478 |
case \[aq]u\[aq]: flags |= CLONE_NEWUTS; break; |
| 1479 |
case \[aq]v\[aq]: verbose = 1; break; |
| 1480 |
case \[aq]z\[aq]: map_zero = 1; break; |
| 1481 |
case \[aq]M\[aq]: uid_map = optarg; break; |
| 1482 |
case \[aq]G\[aq]: gid_map = optarg; break; |
| 1483 |
case \[aq]U\[aq]: flags |= CLONE_NEWUSER; break; |
| 1484 |
default: usage(argv[0]); |
| 1485 |
} |
| 1486 |
} |
| 1487 |
\& |
| 1488 |
/* \-M or \-G without \-U is nonsensical */ |
| 1489 |
\& |
| 1490 |
if (((uid_map != NULL || gid_map != NULL || map_zero) && |
| 1491 |
!(flags & CLONE_NEWUSER)) || |
| 1492 |
(map_zero && (uid_map != NULL || gid_map != NULL))) |
| 1493 |
usage(argv[0]); |
| 1494 |
\& |
| 1495 |
args.argv = &argv[optind]; |
| 1496 |
\& |
| 1497 |
/* We use a pipe to synchronize the parent and child, in order to |
| 1498 |
ensure that the parent sets the UID and GID maps before the child |
| 1499 |
calls execve(). This ensures that the child maintains its |
| 1500 |
capabilities during the execve() in the common case where we |
| 1501 |
want to map the child\[aq]s effective user ID to 0 in the new user |
| 1502 |
namespace. Without this synchronization, the child would lose |
| 1503 |
its capabilities if it performed an execve() with nonzero |
| 1504 |
user IDs (see the capabilities(7) man page for details of the |
| 1505 |
transformation of a process\[aq]s capabilities during execve()). */ |
| 1506 |
\& |
| 1507 |
if (pipe(args.pipe_fd) == \-1) |
| 1508 |
err(EXIT_FAILURE, "pipe"); |
| 1509 |
\& |
| 1510 |
/* Create the child in new namespace(s). */ |
| 1511 |
\& |
| 1512 |
child_pid = clone(childFunc, child_stack + STACK_SIZE, |
| 1513 |
flags | SIGCHLD, &args); |
| 1514 |
if (child_pid == \-1) |
| 1515 |
err(EXIT_FAILURE, "clone"); |
| 1516 |
\& |
| 1517 |
/* Parent falls through to here. */ |
| 1518 |
\& |
| 1519 |
if (verbose) |
| 1520 |
printf("%s: PID of child created by clone() is %jd\[rs]n", |
| 1521 |
argv[0], (intmax_t) child_pid); |
| 1522 |
\& |
| 1523 |
/* Update the UID and GID maps in the child. */ |
| 1524 |
\& |
| 1525 |
if (uid_map != NULL || map_zero) { |
| 1526 |
snprintf(map_path, PATH_MAX, "/proc/%jd/uid_map", |
| 1527 |
(intmax_t) child_pid); |
| 1528 |
if (map_zero) { |
| 1529 |
snprintf(map_buf, MAP_BUF_SIZE, "0 %jd 1", |
| 1530 |
(intmax_t) getuid()); |
| 1531 |
uid_map = map_buf; |
| 1532 |
} |
| 1533 |
update_map(uid_map, map_path); |
| 1534 |
} |
| 1535 |
\& |
| 1536 |
if (gid_map != NULL || map_zero) { |
| 1537 |
proc_setgroups_write(child_pid, "deny"); |
| 1538 |
\& |
| 1539 |
snprintf(map_path, PATH_MAX, "/proc/%jd/gid_map", |
| 1540 |
(intmax_t) child_pid); |
| 1541 |
if (map_zero) { |
| 1542 |
snprintf(map_buf, MAP_BUF_SIZE, "0 %ld 1", |
| 1543 |
(intmax_t) getgid()); |
| 1544 |
gid_map = map_buf; |
| 1545 |
} |
| 1546 |
update_map(gid_map, map_path); |
| 1547 |
} |
| 1548 |
\& |
| 1549 |
/* Close the write end of the pipe, to signal to the child that we |
| 1550 |
have updated the UID and GID maps. */ |
| 1551 |
\& |
| 1552 |
close(args.pipe_fd[1]); |
| 1553 |
\& |
| 1554 |
if (waitpid(child_pid, NULL, 0) == \-1) /* Wait for child */ |
| 1555 |
err(EXIT_FAILURE, "waitpid"); |
| 1556 |
\& |
| 1557 |
if (verbose) |
| 1558 |
printf("%s: terminating\[rs]n", argv[0]); |
| 1559 |
\& |
| 1560 |
exit(EXIT_SUCCESS); |
| 1561 |
} |
| 1562 |
.EE |
| 1563 |
.SH SEE ALSO |
| 1564 |
.BR newgidmap (1), \" From the shadow package |
| 1565 |
.BR newuidmap (1), \" From the shadow package |
| 1566 |
.BR clone (2), |
| 1567 |
.BR ptrace (2), |
| 1568 |
.BR setns (2), |
| 1569 |
.BR unshare (2), |
| 1570 |
.BR proc (5), |
| 1571 |
.BR subgid (5), \" From the shadow package |
| 1572 |
.BR subuid (5), \" From the shadow package |
| 1573 |
.BR capabilities (7), |
| 1574 |
.BR cgroup_namespaces (7), |
| 1575 |
.BR credentials (7), |
| 1576 |
.BR namespaces (7), |
| 1577 |
.BR pid_namespaces (7) |
| 1578 |
.P |
| 1579 |
The kernel source file |
| 1580 |
.IR Documentation/admin\-guide/namespaces/resource\-control.rst . |