man7/pid_namespaces.7

   1 .\" Copyright (c) 2013 by Michael Kerrisk <mtk.manpages@gmail.com>
   2 .\" and Copyright (c) 2012 by Eric W. Biederman <ebiederm@xmission.com>
   3 .\"
   4 .\" %%%LICENSE_START(VERBATIM)
   5 .\" Permission is granted to make and distribute verbatim copies of this
   6 .\" manual provided the copyright notice and this permission notice are
   7 .\" preserved on all copies.
   8 .\"
   9 .\" Permission is granted to copy and distribute modified versions of this
  10 .\" manual under the conditions for verbatim copying, provided that the
  11 .\" entire resulting derived work is distributed under the terms of a
  12 .\" permission notice identical to this one.
  13 .\"
  14 .\" Since the Linux kernel and libraries are constantly changing, this
  15 .\" manual page may be incorrect or out-of-date.  The author(s) assume no
  16 .\" responsibility for errors or omissions, or for damages resulting from
  17 .\" the use of the information contained herein.  The author(s) may not
  18 .\" have taken the same level of care in the production of this manual,
  19 .\" which is licensed free of charge, as they might when working
  20 .\" professionally.
  21 .\"
  22 .\" Formatted or processed versions of this manual, if unaccompanied by
  23 .\" the source, must acknowledge the copyright and authors of this work.
  24 .\" %%%LICENSE_END
  25 .\"
  26 .\"
  27 .TH PID_NAMESPACES 7 2019-03-06 "Linux" "Linux Programmer's Manual"
  28 .SH NAME
  29 pid_namespaces \- overview of Linux PID namespaces
  30 .SH DESCRIPTION
  31 For an overview of namespaces, see
  32 .BR namespaces (7).
  33 .PP
  34 PID namespaces isolate the process ID number space,
  35 meaning that processes in different PID namespaces can have the same PID.
  36 PID namespaces allow containers to provide functionality
  37 such as suspending/resuming the set of processes in the container and
  38 migrating the container to a new host
  39 while the processes inside the container maintain the same PIDs.
  40 .PP
  41 PIDs in a new PID namespace start at 1,
  42 somewhat like a standalone system, and calls to
  43 .BR fork (2),
  44 .BR vfork (2),
  45 or
  46 .BR clone (2)
  47 will produce processes with PIDs that are unique within the namespace.
  48 .PP
  49 Use of PID namespaces requires a kernel that is configured with the
  50 .B CONFIG_PID_NS
  51 option.
  52 .\"
  53 .\" ============================================================
  54 .\"
  55 .SS The namespace "init" process
  56 The first process created in a new namespace
  57 (i.e., the process created using
  58 .BR clone (2)
  59 with the
  60 .BR CLONE_NEWPID
  61 flag, or the first child created by a process after a call to
  62 .BR unshare (2)
  63 using the
  64 .BR CLONE_NEWPID
  65 flag) has the PID 1, and is the "init" process for the namespace (see
  66 .BR init (1)).
  67 This process becomes the parent of any child processes that are orphaned
  68 because a process that resides in this PID namespace terminated
  69 (see below for further details).
  70 .PP
  71 If the "init" process of a PID namespace terminates,
  72 the kernel terminates all of the processes in the namespace via a
  73 .BR SIGKILL
  74 signal.
  75 This behavior reflects the fact that the "init" process
  76 is essential for the correct operation of a PID namespace.
  77 In this case, a subsequent
  78 .BR fork (2)
  79 into this PID namespace fail with the error
  80 .BR ENOMEM ;
  81 it is not possible to create a new process in a PID namespace whose "init"
  82 process has terminated.
  83 Such scenarios can occur when, for example,
  84 a process uses an open file descriptor for a
  85 .I /proc/[pid]/ns/pid
  86 file corresponding to a process that was in a namespace to
  87 .BR setns (2)
  88 into that namespace after the "init" process has terminated.
  89 Another possible scenario can occur after a call to
  90 .BR unshare (2):
  91 if the first child subsequently created by a
  92 .BR fork (2)
  93 terminates, then subsequent calls to
  94 .BR fork (2)
  95 fail with
  96 .BR ENOMEM .
  97 .PP
  98 Only signals for which the "init" process has established a signal handler
  99 can be sent to the "init" process by other members of the PID namespace.
 100 This restriction applies even to privileged processes,
 101 and prevents other members of the PID namespace from
 102 accidentally killing the "init" process.
 103 .PP
 104 Likewise, a process in an ancestor namespace
 105 can\(emsubject to the usual permission checks described in
 106 .BR kill (2)\(emsend
 107 signals to the "init" process of a child PID namespace only
 108 if the "init" process has established a handler for that signal.
 109 (Within the handler, the
 110 .I siginfo_t
 111 .I si_pid
 112 field described in
 113 .BR sigaction (2)
 114 will be zero.)
 115 .B SIGKILL
 116 or
 117 .B SIGSTOP
 118 are treated exceptionally:
 119 these signals are forcibly delivered when sent from an ancestor PID namespace.
 120 Neither of these signals can be caught by the "init" process,
 121 and so will result in the usual actions associated with those signals
 122 (respectively, terminating and stopping the process).
 123 .PP
 124 Starting with Linux 3.4, the
 125 .BR reboot (2)
 126 system call causes a signal to be sent to the namespace "init" process.
 127 See
 128 .BR reboot (2)
 129 for more details.
 130 .\"
 131 .\" ============================================================
 132 .\"
 133 .SS Nesting PID namespaces
 134 PID namespaces can be nested:
 135 each PID namespace has a parent,
 136 except for the initial ("root") PID namespace.
 137 The parent of a PID namespace is the PID namespace of the process that
 138 created the namespace using
 139 .BR clone (2)
 140 or
 141 .BR unshare (2).
 142 PID namespaces thus form a tree,
 143 with all namespaces ultimately tracing their ancestry to the root namespace.
 144 Since Linux 3.7,
 145 .\" commit f2302505775fd13ba93f034206f1e2a587017929
 146 .\" The kernel constant MAX_PID_NS_LEVEL
 147 the kernel limits the maximum nesting depth for PID namespaces to 32.
 148 .PP
 149 A process is visible to other processes in its PID namespace,
 150 and to the processes in each direct ancestor PID namespace
 151 going back to the root PID namespace.
 152 In this context, "visible" means that one process
 153 can be the target of operations by another process using
 154 system calls that specify a process ID.
 155 Conversely, the processes in a child PID namespace can't see
 156 processes in the parent and further removed ancestor namespaces.
 157 More succinctly: a process can see (e.g., send signals with
 158 .BR kill (2),
 159 set nice values with
 160 .BR setpriority (2),
 161 etc.) only processes contained in its own PID namespace
 162 and in descendants of that namespace.
 163 .PP
 164 A process has one process ID in each of the layers of the PID
 165 namespace hierarchy in which is visible,
 166 and walking back though each direct ancestor namespace
 167 through to the root PID namespace.
 168 System calls that operate on process IDs always
 169 operate using the process ID that is visible in the
 170 PID namespace of the caller.
 171 A call to
 172 .BR getpid (2)
 173 always returns the PID associated with the namespace in which
 174 the process was created.
 175 .PP
 176 Some processes in a PID namespace may have parents
 177 that are outside of the namespace.
 178 For example, the parent of the initial process in the namespace
 179 (i.e., the
 180 .BR init (1)
 181 process with PID 1) is necessarily in another namespace.
 182 Likewise, the direct children of a process that uses
 183 .BR setns (2)
 184 to cause its children to join a PID namespace are in a different
 185 PID namespace from the caller of
 186 .BR setns (2).
 187 Calls to
 188 .BR getppid (2)
 189 for such processes return 0.
 190 .PP
 191 While processes may freely descend into child PID namespaces
 192 (e.g., using
 193 .BR setns (2)
 194 with a PID namespace file descriptor),
 195 they may not move in the other direction.
 196 That is to say, processes may not enter any ancestor namespaces
 197 (parent, grandparent, etc.).
 198 Changing PID namespaces is a one-way operation.
 199 .PP
 200 The
 201 .BR NS_GET_PARENT
 202 .BR ioctl (2)
 203 operation can be used to discover the parental relationship
 204 between PID namespaces; see
 205 .BR ioctl_ns (2).
 206 .\"
 207 .\" ============================================================
 208 .\"
 209 .SS setns(2) and unshare(2) semantics
 210 Calls to
 211 .BR setns (2)
 212 that specify a PID namespace file descriptor
 213 and calls to
 214 .BR unshare (2)
 215 with the
 216 .BR CLONE_NEWPID
 217 flag cause children subsequently created
 218 by the caller to be placed in a different PID namespace from the caller.
 219 (Since Linux 4.12, that PID namespace is shown via the
 220 .IR /proc/[pid]/ns/pid_for_children
 221 file, as described in
 222 .BR namespaces (7).)
 223 These calls do not, however,
 224 change the PID namespace of the calling process,
 225 because doing so would change the caller's idea of its own PID
 226 (as reported by
 227 .BR getpid ()),
 228 which would break many applications and libraries.
 229 .PP
 230 To put things another way:
 231 a process's PID namespace membership is determined when the process is created
 232 and cannot be changed thereafter.
 233 Among other things, this means that the parental relationship
 234 between processes mirrors the parental relationship between PID namespaces:
 235 the parent of a process is either in the same namespace
 236 or resides in the immediate parent PID namespace.
 237 .PP
 238 A process may call
 239 .BR unshare (2)
 240 with the
 241 .B CLONE_NEWPID
 242 flag only once.
 243 After it has performed this operation, its
 244 .IR /proc/PID/ns/pid_for_children
 245 symbolic link will be empty until the first child is created in the namespace.
 246 .\"
 247 .\" ============================================================
 248 .\"
 249 .SS Adoption of orphaned children
 250 When a child process becomes orphaned, it is reparented to the "init"
 251 process in the PID namespace of its parent
 252 (unless one of the nearer ancestors of the parent employed the
 253 .BR prctl (2)
 254 .B PR_SET_CHILD_SUBREAPER
 255 command to mark itself as the reaper of orphaned descendant processes).
 256 Note that because of the
 257 .BR setns (2)
 258 and
 259 .BR unshare (2)
 260 semantics described above, this may be the "init" process in the PID
 261 namespace that is the
 262 .I parent
 263 of the child's PID namespace,
 264 rather than the "init" process in the child's own PID namespace.
 265 \" Furthermore, by definition, the parent of the "init" process
 266 .\" of a PID namespace resides in the parent PID namespace.
 267 .\"
 268 .\" ============================================================
 269 .\"
 270 .SS Compatibility of CLONE_NEWPID with other CLONE_* flags
 271 In current versions of Linux,
 272 .BR CLONE_NEWPID
 273 can't be combined with
 274 .BR CLONE_THREAD .
 275 Threads are required to be in the same PID namespace such that
 276 the threads in a process can send signals to each other.
 277 Similarly, it must be possible to see all of the threads
 278 of a processes in the
 279 .BR proc (5)
 280 filesystem.
 281 Additionally, if two threads were in different PID
 282 namespaces, the process ID of the process sending a signal
 283 could not be meaningfully encoded when a signal is sent
 284 (see the description of the
 285 .I siginfo_t
 286 type in
 287 .BR sigaction (2)).
 288 Since this is computed when a signal is enqueued,
 289 a signal queue shared by processes in multiple PID namespaces
 290 would defeat that.
 291 .PP
 292 .\" Note these restrictions were all introduced in
 293 .\" 8382fcac1b813ad0a4e68a838fc7ae93fa39eda0
 294 .\" when CLONE_NEWPID|CLONE_VM was disallowed
 295 In earlier versions of Linux,
 296 .BR CLONE_NEWPID
 297 was additionally disallowed (failing with the error
 298 .BR EINVAL )
 299 in combination with
 300 .BR CLONE_SIGHAND
 301 .\" (restriction lifted in faf00da544045fdc1454f3b9e6d7f65c841de302)
 302 (before Linux 4.3) as well as
 303 .\" (restriction lifted in e79f525e99b04390ca4d2366309545a836c03bf1)
 304 .BR CLONE_VM
 305 (before Linux 3.12).
 306 The changes that lifted these restrictions have also been ported to
 307 earlier stable kernels.
 308 .\"
 309 .\" ============================================================
 310 .\"
 311 .SS /proc and PID namespaces
 312 A
 313 .I /proc
 314 filesystem shows (in the
 315 .I /proc/[pid]
 316 directories) only processes visible in the PID namespace
 317 of the process that performed the mount, even if the
 318 .I /proc
 319 filesystem is viewed from processes in other namespaces.
 320 .PP
 321 After creating a new PID namespace,
 322 it is useful for the child to change its root directory
 323 and mount a new procfs instance at
 324 .I /proc
 325 so that tools such as
 326 .BR ps (1)
 327 work correctly.
 328 If a new mount namespace is simultaneously created by including
 329 .BR CLONE_NEWNS
 330 in the
 331 .IR flags
 332 argument of
 333 .BR clone (2)
 334 or
 335 .BR unshare (2),
 336 then it isn't necessary to change the root directory:
 337 a new procfs instance can be mounted directly over
 338 .IR /proc .
 339 .PP
 340 From a shell, the command to mount
 341 .I /proc
 342 is:
 343 .PP
 344 .in +4n
 345 .EX
 346 $ mount -t proc proc /proc
 347 .EE
 348 .in
 349 .PP
 350 Calling
 351 .BR readlink (2)
 352 on the path
 353 .I /proc/self
 354 yields the process ID of the caller in the PID namespace of the procfs mount
 355 (i.e., the PID namespace of the process that mounted the procfs).
 356 This can be useful for introspection purposes,
 357 when a process wants to discover its PID in other namespaces.
 358 .\"
 359 .\" ============================================================
 360 .\"
 361 .SS /proc files
 362 .TP
 363 .BR /proc/sys/kernel/ns_last_pid " (since Linux 3.3)"
 364 .\" commit b8f566b04d3cddd192cfd2418ae6d54ac6353792
 365 This file
 366 (which is virtualized per PID namespace)
 367 displays the last PID that was allocated in this PID namespace.
 368 When the next PID is allocated,
 369 the kernel will search for the lowest unallocated PID
 370 that is greater than this value,
 371 and when this file is subsequently read it will show that PID.
 372 .IP
 373 This file is writable by a process that has the
 374 .B CAP_SYS_ADMIN
 375 capability inside the user namespace that owns the PID namespace.
 376 .\" This ability is necessary to support checkpoint restore in user-space
 377 This makes it possible to determine the PID that is allocated
 378 to the next process that is created inside this PID namespace.
 379 .\"
 380 .\" ============================================================
 381 .\"
 382 .SS Miscellaneous
 383 When a process ID is passed over a UNIX domain socket to a
 384 process in a different PID namespace (see the description of
 385 .B SCM_CREDENTIALS
 386 in
 387 .BR unix (7)),
 388 it is translated into the corresponding PID value in
 389 the receiving process's PID namespace.
 390 .SH CONFORMING TO
 391 Namespaces are a Linux-specific feature.
 392 .SH EXAMPLE
 393 See
 394 .BR user_namespaces (7).
 395 .SH SEE ALSO
 396 .BR clone (2),
 397 .BR reboot (2),
 398 .BR setns (2),
 399 .BR unshare (2),
 400 .BR proc (5),
 401 .BR capabilities (7),
 402 .BR credentials (7),
 403 .BR mount_namespaces (7),
 404 .BR namespaces (7),
 405 .BR user_namespaces (7),
 406 .BR switch_root (8)