1CGROUPS(7) Linux Programmer's Manual CGROUPS(7)
2
3
4
6 cgroups - Linux control groups
7
9 Control groups, usually referred to as cgroups, are a Linux kernel fea‐
10 ture which allow processes to be organized into hierarchical groups
11 whose usage of various types of resources can then be limited and moni‐
12 tored. The kernel's cgroup interface is provided through a pseudo-
13 filesystem called cgroupfs. Grouping is implemented in the core cgroup
14 kernel code, while resource tracking and limits are implemented in a
15 set of per-resource-type subsystems (memory, CPU, and so on).
16
17 Terminology
18 A cgroup is a collection of processes that are bound to a set of limits
19 or parameters defined via the cgroup filesystem.
20
21 A subsystem is a kernel component that modifies the behavior of the
22 processes in a cgroup. Various subsystems have been implemented, mak‐
23 ing it possible to do things such as limiting the amount of CPU time
24 and memory available to a cgroup, accounting for the CPU time used by a
25 cgroup, and freezing and resuming execution of the processes in a
26 cgroup. Subsystems are sometimes also known as resource controllers
27 (or simply, controllers).
28
29 The cgroups for a controller are arranged in a hierarchy. This hierar‐
30 chy is defined by creating, removing, and renaming subdirectories
31 within the cgroup filesystem. At each level of the hierarchy, at‐
32 tributes (e.g., limits) can be defined. The limits, control, and ac‐
33 counting provided by cgroups generally have effect throughout the sub‐
34 hierarchy underneath the cgroup where the attributes are defined.
35 Thus, for example, the limits placed on a cgroup at a higher level in
36 the hierarchy cannot be exceeded by descendant cgroups.
37
38 Cgroups version 1 and version 2
39 The initial release of the cgroups implementation was in Linux 2.6.24.
40 Over time, various cgroup controllers have been added to allow the man‐
41 agement of various types of resources. However, the development of
42 these controllers was largely uncoordinated, with the result that many
43 inconsistencies arose between controllers and management of the cgroup
44 hierarchies became rather complex. A longer description of these prob‐
45 lems can be found in the kernel source file Documentation/ad‐
46 min-guide/cgroup-v2.rst (or Documentation/cgroup-v2.txt in Linux 4.17
47 and earlier).
48
49 Because of the problems with the initial cgroups implementation
50 (cgroups version 1), starting in Linux 3.10, work began on a new, or‐
51 thogonal implementation to remedy these problems. Initially marked ex‐
52 perimental, and hidden behind the -o __DEVEL__sane_behavior mount op‐
53 tion, the new version (cgroups version 2) was eventually made official
54 with the release of Linux 4.5. Differences between the two versions
55 are described in the text below. The file cgroup.sane_behavior,
56 present in cgroups v1, is a relic of this mount option. The file al‐
57 ways reports "0" and is only retained for backward compatibility.
58
59 Although cgroups v2 is intended as a replacement for cgroups v1, the
60 older system continues to exist (and for compatibility reasons is un‐
61 likely to be removed). Currently, cgroups v2 implements only a subset
62 of the controllers available in cgroups v1. The two systems are imple‐
63 mented so that both v1 controllers and v2 controllers can be mounted on
64 the same system. Thus, for example, it is possible to use those con‐
65 trollers that are supported under version 2, while also using version 1
66 controllers where version 2 does not yet support those controllers.
67 The only restriction here is that a controller can't be simultaneously
68 employed in both a cgroups v1 hierarchy and in the cgroups v2 hierar‐
69 chy.
70
72 Under cgroups v1, each controller may be mounted against a separate
73 cgroup filesystem that provides its own hierarchical organization of
74 the processes on the system. It is also possible to comount multiple
75 (or even all) cgroups v1 controllers against the same cgroup filesys‐
76 tem, meaning that the comounted controllers manage the same hierarchi‐
77 cal organization of processes.
78
79 For each mounted hierarchy, the directory tree mirrors the control
80 group hierarchy. Each control group is represented by a directory,
81 with each of its child control cgroups represented as a child direc‐
82 tory. For instance, /user/joe/1.session represents control group
83 1.session, which is a child of cgroup joe, which is a child of /user.
84 Under each cgroup directory is a set of files which can be read or
85 written to, reflecting resource limits and a few general cgroup proper‐
86 ties.
87
88 Tasks (threads) versus processes
89 In cgroups v1, a distinction is drawn between processes and tasks. In
90 this view, a process can consist of multiple tasks (more commonly
91 called threads, from a user-space perspective, and called such in the
92 remainder of this man page). In cgroups v1, it is possible to indepen‐
93 dently manipulate the cgroup memberships of the threads in a process.
94
95 The cgroups v1 ability to split threads across different cgroups caused
96 problems in some cases. For example, it made no sense for the memory
97 controller, since all of the threads of a process share a single ad‐
98 dress space. Because of these problems, the ability to independently
99 manipulate the cgroup memberships of the threads in a process was re‐
100 moved in the initial cgroups v2 implementation, and subsequently re‐
101 stored in a more limited form (see the discussion of "thread mode" be‐
102 low).
103
104 Mounting v1 controllers
105 The use of cgroups requires a kernel built with the CONFIG_CGROUP op‐
106 tion. In addition, each of the v1 controllers has an associated con‐
107 figuration option that must be set in order to employ that controller.
108
109 In order to use a v1 controller, it must be mounted against a cgroup
110 filesystem. The usual place for such mounts is under a tmpfs(5)
111 filesystem mounted at /sys/fs/cgroup. Thus, one might mount the cpu
112 controller as follows:
113
114 mount -t cgroup -o cpu none /sys/fs/cgroup/cpu
115
116 It is possible to comount multiple controllers against the same hierar‐
117 chy. For example, here the cpu and cpuacct controllers are comounted
118 against a single hierarchy:
119
120 mount -t cgroup -o cpu,cpuacct none /sys/fs/cgroup/cpu,cpuacct
121
122 Comounting controllers has the effect that a process is in the same
123 cgroup for all of the comounted controllers. Separately mounting con‐
124 trollers allows a process to be in cgroup /foo1 for one controller
125 while being in /foo2/foo3 for another.
126
127 It is possible to comount all v1 controllers against the same hierar‐
128 chy:
129
130 mount -t cgroup -o all cgroup /sys/fs/cgroup
131
132 (One can achieve the same result by omitting -o all, since it is the
133 default if no controllers are explicitly specified.)
134
135 It is not possible to mount the same controller against multiple cgroup
136 hierarchies. For example, it is not possible to mount both the cpu and
137 cpuacct controllers against one hierarchy, and to mount the cpu con‐
138 troller alone against another hierarchy. It is possible to create mul‐
139 tiple mount with exactly the same set of comounted controllers. How‐
140 ever, in this case all that results is multiple mount points providing
141 a view of the same hierarchy.
142
143 Note that on many systems, the v1 controllers are automatically mounted
144 under /sys/fs/cgroup; in particular, systemd(1) automatically creates
145 such mounts.
146
147 Unmounting v1 controllers
148 A mounted cgroup filesystem can be unmounted using the umount(8) com‐
149 mand, as in the following example:
150
151 umount /sys/fs/cgroup/pids
152
153 But note well: a cgroup filesystem is unmounted only if it is not busy,
154 that is, it has no child cgroups. If this is not the case, then the
155 only effect of the umount(8) is to make the mount invisible. Thus, to
156 ensure that the mount is really removed, one must first remove all
157 child cgroups, which in turn can be done only after all member pro‐
158 cesses have been moved from those cgroups to the root cgroup.
159
160 Cgroups version 1 controllers
161 Each of the cgroups version 1 controllers is governed by a kernel con‐
162 figuration option (listed below). Additionally, the availability of
163 the cgroups feature is governed by the CONFIG_CGROUPS kernel configura‐
164 tion option.
165
166 cpu (since Linux 2.6.24; CONFIG_CGROUP_SCHED)
167 Cgroups can be guaranteed a minimum number of "CPU shares" when
168 a system is busy. This does not limit a cgroup's CPU usage if
169 the CPUs are not busy. For further information, see Documenta‐
170 tion/scheduler/sched-design-CFS.rst (or Documentation/sched‐
171 uler/sched-design-CFS.txt in Linux 5.2 and earlier).
172
173 In Linux 3.2, this controller was extended to provide CPU "band‐
174 width" control. If the kernel is configured with CON‐
175 FIG_CFS_BANDWIDTH, then within each scheduling period (defined
176 via a file in the cgroup directory), it is possible to define an
177 upper limit on the CPU time allocated to the processes in a
178 cgroup. This upper limit applies even if there is no other com‐
179 petition for the CPU. Further information can be found in the
180 kernel source file Documentation/scheduler/sched-bwc.rst (or
181 Documentation/scheduler/sched-bwc.txt in Linux 5.2 and earlier).
182
183 cpuacct (since Linux 2.6.24; CONFIG_CGROUP_CPUACCT)
184 This provides accounting for CPU usage by groups of processes.
185
186 Further information can be found in the kernel source file Docu‐
187 mentation/admin-guide/cgroup-v1/cpuacct.rst (or Documenta‐
188 tion/cgroup-v1/cpuacct.txt in Linux 5.2 and earlier).
189
190 cpuset (since Linux 2.6.24; CONFIG_CPUSETS)
191 This cgroup can be used to bind the processes in a cgroup to a
192 specified set of CPUs and NUMA nodes.
193
194 Further information can be found in the kernel source file Docu‐
195 mentation/admin-guide/cgroup-v1/cpusets.rst (or Documenta‐
196 tion/cgroup-v1/cpusets.txt in Linux 5.2 and earlier).
197
198 memory (since Linux 2.6.25; CONFIG_MEMCG)
199 The memory controller supports reporting and limiting of process
200 memory, kernel memory, and swap used by cgroups.
201
202 Further information can be found in the kernel source file Docu‐
203 mentation/admin-guide/cgroup-v1/memory.rst (or Documenta‐
204 tion/cgroup-v1/memory.txt in Linux 5.2 and earlier).
205
206 devices (since Linux 2.6.26; CONFIG_CGROUP_DEVICE)
207 This supports controlling which processes may create (mknod) de‐
208 vices as well as open them for reading or writing. The policies
209 may be specified as allow-lists and deny-lists. Hierarchy is
210 enforced, so new rules must not violate existing rules for the
211 target or ancestor cgroups.
212
213 Further information can be found in the kernel source file Docu‐
214 mentation/admin-guide/cgroup-v1/devices.rst (or Documenta‐
215 tion/cgroup-v1/devices.txt in Linux 5.2 and earlier).
216
217 freezer (since Linux 2.6.28; CONFIG_CGROUP_FREEZER)
218 The freezer cgroup can suspend and restore (resume) all pro‐
219 cesses in a cgroup. Freezing a cgroup /A also causes its chil‐
220 dren, for example, processes in /A/B, to be frozen.
221
222 Further information can be found in the kernel source file Docu‐
223 mentation/admin-guide/cgroup-v1/freezer-subsystem.rst (or Docu‐
224 mentation/cgroup-v1/freezer-subsystem.txt in Linux 5.2 and ear‐
225 lier).
226
227 net_cls (since Linux 2.6.29; CONFIG_CGROUP_NET_CLASSID)
228 This places a classid, specified for the cgroup, on network
229 packets created by a cgroup. These classids can then be used in
230 firewall rules, as well as used to shape traffic using tc(8).
231 This applies only to packets leaving the cgroup, not to traffic
232 arriving at the cgroup.
233
234 Further information can be found in the kernel source file Docu‐
235 mentation/admin-guide/cgroup-v1/net_cls.rst (or Documenta‐
236 tion/cgroup-v1/net_cls.txt in Linux 5.2 and earlier).
237
238 blkio (since Linux 2.6.33; CONFIG_BLK_CGROUP)
239 The blkio cgroup controls and limits access to specified block
240 devices by applying IO control in the form of throttling and up‐
241 per limits against leaf nodes and intermediate nodes in the
242 storage hierarchy.
243
244 Two policies are available. The first is a proportional-weight
245 time-based division of disk implemented with CFQ. This is in
246 effect for leaf nodes using CFQ. The second is a throttling
247 policy which specifies upper I/O rate limits on a device.
248
249 Further information can be found in the kernel source file Docu‐
250 mentation/admin-guide/cgroup-v1/blkio-controller.rst (or Docu‐
251 mentation/cgroup-v1/blkio-controller.txt in Linux 5.2 and ear‐
252 lier).
253
254 perf_event (since Linux 2.6.39; CONFIG_CGROUP_PERF)
255 This controller allows perf monitoring of the set of processes
256 grouped in a cgroup.
257
258 Further information can be found in the kernel source files
259
260 net_prio (since Linux 3.3; CONFIG_CGROUP_NET_PRIO)
261 This allows priorities to be specified, per network interface,
262 for cgroups.
263
264 Further information can be found in the kernel source file Docu‐
265 mentation/admin-guide/cgroup-v1/net_prio.rst (or Documenta‐
266 tion/cgroup-v1/net_prio.txt in Linux 5.2 and earlier).
267
268 hugetlb (since Linux 3.5; CONFIG_CGROUP_HUGETLB)
269 This supports limiting the use of huge pages by cgroups.
270
271 Further information can be found in the kernel source file Docu‐
272 mentation/admin-guide/cgroup-v1/hugetlb.rst (or Documenta‐
273 tion/cgroup-v1/hugetlb.txt in Linux 5.2 and earlier).
274
275 pids (since Linux 4.3; CONFIG_CGROUP_PIDS)
276 This controller permits limiting the number of process that may
277 be created in a cgroup (and its descendants).
278
279 Further information can be found in the kernel source file Docu‐
280 mentation/admin-guide/cgroup-v1/pids.rst (or Documenta‐
281 tion/cgroup-v1/pids.txt in Linux 5.2 and earlier).
282
283 rdma (since Linux 4.11; CONFIG_CGROUP_RDMA)
284 The RDMA controller permits limiting the use of RDMA/IB-specific
285 resources per cgroup.
286
287 Further information can be found in the kernel source file Docu‐
288 mentation/admin-guide/cgroup-v1/rdma.rst (or Documenta‐
289 tion/cgroup-v1/rdma.txt in Linux 5.2 and earlier).
290
291 Creating cgroups and moving processes
292 A cgroup filesystem initially contains a single root cgroup, '/', which
293 all processes belong to. A new cgroup is created by creating a direc‐
294 tory in the cgroup filesystem:
295
296 mkdir /sys/fs/cgroup/cpu/cg1
297
298 This creates a new empty cgroup.
299
300 A process may be moved to this cgroup by writing its PID into the
301 cgroup's cgroup.procs file:
302
303 echo $$ > /sys/fs/cgroup/cpu/cg1/cgroup.procs
304
305 Only one PID at a time should be written to this file.
306
307 Writing the value 0 to a cgroup.procs file causes the writing process
308 to be moved to the corresponding cgroup.
309
310 When writing a PID into the cgroup.procs, all threads in the process
311 are moved into the new cgroup at once.
312
313 Within a hierarchy, a process can be a member of exactly one cgroup.
314 Writing a process's PID to a cgroup.procs file automatically removes it
315 from the cgroup of which it was previously a member.
316
317 The cgroup.procs file can be read to obtain a list of the processes
318 that are members of a cgroup. The returned list of PIDs is not guaran‐
319 teed to be in order. Nor is it guaranteed to be free of duplicates.
320 (For example, a PID may be recycled while reading from the list.)
321
322 In cgroups v1, an individual thread can be moved to another cgroup by
323 writing its thread ID (i.e., the kernel thread ID returned by clone(2)
324 and gettid(2)) to the tasks file in a cgroup directory. This file can
325 be read to discover the set of threads that are members of the cgroup.
326
327 Removing cgroups
328 To remove a cgroup, it must first have no child cgroups and contain no
329 (nonzombie) processes. So long as that is the case, one can simply re‐
330 move the corresponding directory pathname. Note that files in a cgroup
331 directory cannot and need not be removed.
332
333 Cgroups v1 release notification
334 Two files can be used to determine whether the kernel provides notifi‐
335 cations when a cgroup becomes empty. A cgroup is considered to be
336 empty when it contains no child cgroups and no member processes.
337
338 A special file in the root directory of each cgroup hierarchy, re‐
339 lease_agent, can be used to register the pathname of a program that may
340 be invoked when a cgroup in the hierarchy becomes empty. The pathname
341 of the newly empty cgroup (relative to the cgroup mount point) is pro‐
342 vided as the sole command-line argument when the release_agent program
343 is invoked. The release_agent program might remove the cgroup direc‐
344 tory, or perhaps repopulate it with a process.
345
346 The default value of the release_agent file is empty, meaning that no
347 release agent is invoked.
348
349 The content of the release_agent file can also be specified via a mount
350 option when the cgroup filesystem is mounted:
351
352 mount -o release_agent=pathname ...
353
354 Whether or not the release_agent program is invoked when a particular
355 cgroup becomes empty is determined by the value in the notify_on_re‐
356 lease file in the corresponding cgroup directory. If this file con‐
357 tains the value 0, then the release_agent program is not invoked. If
358 it contains the value 1, the release_agent program is invoked. The de‐
359 fault value for this file in the root cgroup is 0. At the time when a
360 new cgroup is created, the value in this file is inherited from the
361 corresponding file in the parent cgroup.
362
363 Cgroup v1 named hierarchies
364 In cgroups v1, it is possible to mount a cgroup hierarchy that has no
365 attached controllers:
366
367 mount -t cgroup -o none,name=somename none /some/mount/point
368
369 Multiple instances of such hierarchies can be mounted; each hierarchy
370 must have a unique name. The only purpose of such hierarchies is to
371 track processes. (See the discussion of release notification below.)
372 An example of this is the name=systemd cgroup hierarchy that is used by
373 systemd(1) to track services and user sessions.
374
375 Since Linux 5.0, the cgroup_no_v1 kernel boot option (described below)
376 can be used to disable cgroup v1 named hierarchies, by specifying
377 cgroup_no_v1=named.
378
380 In cgroups v2, all mounted controllers reside in a single unified hier‐
381 archy. While (different) controllers may be simultaneously mounted un‐
382 der the v1 and v2 hierarchies, it is not possible to mount the same
383 controller simultaneously under both the v1 and the v2 hierarchies.
384
385 The new behaviors in cgroups v2 are summarized here, and in some cases
386 elaborated in the following subsections.
387
388 1. Cgroups v2 provides a unified hierarchy against which all con‐
389 trollers are mounted.
390
391 2. "Internal" processes are not permitted. With the exception of the
392 root cgroup, processes may reside only in leaf nodes (cgroups that
393 do not themselves contain child cgroups). The details are somewhat
394 more subtle than this, and are described below.
395
396 3. Active cgroups must be specified via the files cgroup.controllers
397 and cgroup.subtree_control.
398
399 4. The tasks file has been removed. In addition, the
400 cgroup.clone_children file that is employed by the cpuset controller
401 has been removed.
402
403 5. An improved mechanism for notification of empty cgroups is provided
404 by the cgroup.events file.
405
406 For more changes, see the Documentation/admin-guide/cgroup-v2.rst file
407 in the kernel source (or Documentation/cgroup-v2.txt in Linux 4.17 and
408 earlier).
409
410 Some of the new behaviors listed above saw subsequent modification with
411 the addition in Linux 4.14 of "thread mode" (described below).
412
413 Cgroups v2 unified hierarchy
414 In cgroups v1, the ability to mount different controllers against dif‐
415 ferent hierarchies was intended to allow great flexibility for applica‐
416 tion design. In practice, though, the flexibility turned out to be
417 less useful than expected, and in many cases added complexity. There‐
418 fore, in cgroups v2, all available controllers are mounted against a
419 single hierarchy. The available controllers are automatically mounted,
420 meaning that it is not necessary (or possible) to specify the con‐
421 trollers when mounting the cgroup v2 filesystem using a command such as
422 the following:
423
424 mount -t cgroup2 none /mnt/cgroup2
425
426 A cgroup v2 controller is available only if it is not currently in use
427 via a mount against a cgroup v1 hierarchy. Or, to put things another
428 way, it is not possible to employ the same controller against both a v1
429 hierarchy and the unified v2 hierarchy. This means that it may be nec‐
430 essary first to unmount a v1 controller (as described above) before
431 that controller is available in v2. Since systemd(1) makes heavy use
432 of some v1 controllers by default, it can in some cases be simpler to
433 boot the system with selected v1 controllers disabled. To do this,
434 specify the cgroup_no_v1=list option on the kernel boot command line;
435 list is a comma-separated list of the names of the controllers to dis‐
436 able, or the word all to disable all v1 controllers. (This situation
437 is correctly handled by systemd(1), which falls back to operating with‐
438 out the specified controllers.)
439
440 Note that on many modern systems, systemd(1) automatically mounts the
441 cgroup2 filesystem at /sys/fs/cgroup/unified during the boot process.
442
443 Cgroups v2 mount options
444 The following options (mount -o) can be specified when mounting the
445 group v2 filesystem:
446
447 nsdelegate (since Linux 4.15)
448 Treat cgroup namespaces as delegation boundaries. For details,
449 see below.
450
451 memory_localevents (since Linux 5.2)
452 The memory.events should show statistics only for the cgroup it‐
453 self, and not for any descendant cgroups. This was the behavior
454 before Linux 5.2. Starting in Linux 5.2, the default behavior
455 is to include statistics for descendant cgroups in mem‐
456 ory.events, and this mount option can be used to revert to the
457 legacy behavior. This option is system wide and can be set on
458 mount or modified through remount only from the initial mount
459 namespace; it is silently ignored in noninitial namespaces.
460
461 Cgroups v2 controllers
462 The following controllers, documented in the kernel source file Docu‐
463 mentation/admin-guide/cgroup-v2.rst (or Documentation/cgroup-v2.txt in
464 Linux 4.17 and earlier), are supported in cgroups version 2:
465
466 cpu (since Linux 4.15)
467 This is the successor to the version 1 cpu and cpuacct con‐
468 trollers.
469
470 cpuset (since Linux 5.0)
471 This is the successor of the version 1 cpuset controller.
472
473 freezer (since Linux 5.2)
474 This is the successor of the version 1 freezer controller.
475
476 hugetlb (since Linux 5.6)
477 This is the successor of the version 1 hugetlb controller.
478
479 io (since Linux 4.5)
480 This is the successor of the version 1 blkio controller.
481
482 memory (since Linux 4.5)
483 This is the successor of the version 1 memory controller.
484
485 perf_event (since Linux 4.11)
486 This is the same as the version 1 perf_event controller.
487
488 pids (since Linux 4.5)
489 This is the same as the version 1 pids controller.
490
491 rdma (since Linux 4.11)
492 This is the same as the version 1 rdma controller.
493
494 There is no direct equivalent of the net_cls and net_prio controllers
495 from cgroups version 1. Instead, support has been added to iptables(8)
496 to allow eBPF filters that hook on cgroup v2 pathnames to make deci‐
497 sions about network traffic on a per-cgroup basis.
498
499 The v2 devices controller provides no interface files; instead, device
500 control is gated by attaching an eBPF (BPF_CGROUP_DEVICE) program to a
501 v2 cgroup.
502
503 Cgroups v2 subtree control
504 Each cgroup in the v2 hierarchy contains the following two files:
505
506 cgroup.controllers
507 This read-only file exposes a list of the controllers that are
508 available in this cgroup. The contents of this file match the
509 contents of the cgroup.subtree_control file in the parent
510 cgroup.
511
512 cgroup.subtree_control
513 This is a list of controllers that are active (enabled) in the
514 cgroup. The set of controllers in this file is a subset of the
515 set in the cgroup.controllers of this cgroup. The set of active
516 controllers is modified by writing strings to this file contain‐
517 ing space-delimited controller names, each preceded by '+' (to
518 enable a controller) or '-' (to disable a controller), as in the
519 following example:
520
521 echo '+pids -memory' > x/y/cgroup.subtree_control
522
523 An attempt to enable a controller that is not present in
524 cgroup.controllers leads to an ENOENT error when writing to the
525 cgroup.subtree_control file.
526
527 Because the list of controllers in cgroup.subtree_control is a subset
528 of those cgroup.controllers, a controller that has been disabled in one
529 cgroup in the hierarchy can never be re-enabled in the subtree below
530 that cgroup.
531
532 A cgroup's cgroup.subtree_control file determines the set of con‐
533 trollers that are exercised in the child cgroups. When a controller
534 (e.g., pids) is present in the cgroup.subtree_control file of a parent
535 cgroup, then the corresponding controller-interface files (e.g.,
536 pids.max) are automatically created in the children of that cgroup and
537 can be used to exert resource control in the child cgroups.
538
539 Cgroups v2 "no internal processes" rule
540 Cgroups v2 enforces a so-called "no internal processes" rule. Roughly
541 speaking, this rule means that, with the exception of the root cgroup,
542 processes may reside only in leaf nodes (cgroups that do not themselves
543 contain child cgroups). This avoids the need to decide how to parti‐
544 tion resources between processes which are members of cgroup A and pro‐
545 cesses in child cgroups of A.
546
547 For instance, if cgroup /cg1/cg2 exists, then a process may reside in
548 /cg1/cg2, but not in /cg1. This is to avoid an ambiguity in cgroups v1
549 with respect to the delegation of resources between processes in /cg1
550 and its child cgroups. The recommended approach in cgroups v2 is to
551 create a subdirectory called leaf for any nonleaf cgroup which should
552 contain processes, but no child cgroups. Thus, processes which previ‐
553 ously would have gone into /cg1 would now go into /cg1/leaf. This has
554 the advantage of making explicit the relationship between processes in
555 /cg1/leaf and /cg1's other children.
556
557 The "no internal processes" rule is in fact more subtle than stated
558 above. More precisely, the rule is that a (nonroot) cgroup can't both
559 (1) have member processes, and (2) distribute resources into child
560 cgroups—that is, have a nonempty cgroup.subtree_control file. Thus, it
561 is possible for a cgroup to have both member processes and child
562 cgroups, but before controllers can be enabled for that cgroup, the
563 member processes must be moved out of the cgroup (e.g., perhaps into
564 the child cgroups).
565
566 With the Linux 4.14 addition of "thread mode" (described below), the
567 "no internal processes" rule has been relaxed in some cases.
568
569 Cgroups v2 cgroup.events file
570 Each nonroot cgroup in the v2 hierarchy contains a read-only file,
571 cgroup.events, whose contents are key-value pairs (delimited by newline
572 characters, with the key and value separated by spaces) providing state
573 information about the cgroup:
574
575 $ cat mygrp/cgroup.events
576 populated 1
577 frozen 0
578
579 The following keys may appear in this file:
580
581 populated
582 The value of this key is either 1, if this cgroup or any of its
583 descendants has member processes, or otherwise 0.
584
585 frozen (since Linux 5.2)
586 The value of this key is 1 if this cgroup is currently frozen,
587 or 0 if it is not.
588
589 The cgroup.events file can be monitored, in order to receive notifica‐
590 tion when the value of one of its keys changes. Such monitoring can be
591 done using inotify(7), which notifies changes as IN_MODIFY events, or
592 poll(2), which notifies changes by returning the POLLPRI and POLLERR
593 bits in the revents field.
594
595 Cgroup v2 release notification
596 Cgroups v2 provides a new mechanism for obtaining notification when a
597 cgroup becomes empty. The cgroups v1 release_agent and notify_on_re‐
598 lease files are removed, and replaced by the populated key in the
599 cgroup.events file. This key either has the value 0, meaning that the
600 cgroup (and its descendants) contain no (nonzombie) member processes,
601 or 1, meaning that the cgroup (or one of its descendants) contains mem‐
602 ber processes.
603
604 The cgroups v2 release-notification mechanism offers the following ad‐
605 vantages over the cgroups v1 release_agent mechanism:
606
607 * It allows for cheaper notification, since a single process can moni‐
608 tor multiple cgroup.events files (using the techniques described
609 earlier). By contrast, the cgroups v1 mechanism requires the ex‐
610 pense of creating a process for each notification.
611
612 * Notification for different cgroup subhierarchies can be delegated to
613 different processes. By contrast, the cgroups v1 mechanism allows
614 only one release agent for an entire hierarchy.
615
616 Cgroups v2 cgroup.stat file
617 Each cgroup in the v2 hierarchy contains a read-only cgroup.stat file
618 (first introduced in Linux 4.14) that consists of lines containing key-
619 value pairs. The following keys currently appear in this file:
620
621 nr_descendants
622 This is the total number of visible (i.e., living) descendant
623 cgroups underneath this cgroup.
624
625 nr_dying_descendants
626 This is the total number of dying descendant cgroups underneath
627 this cgroup. A cgroup enters the dying state after being
628 deleted. It remains in that state for an undefined period
629 (which will depend on system load) while resources are freed be‐
630 fore the cgroup is destroyed. Note that the presence of some
631 cgroups in the dying state is normal, and is not indicative of
632 any problem.
633
634 A process can't be made a member of a dying cgroup, and a dying
635 cgroup can't be brought back to life.
636
637 Limiting the number of descendant cgroups
638 Each cgroup in the v2 hierarchy contains the following files, which can
639 be used to view and set limits on the number of descendant cgroups un‐
640 der that cgroup:
641
642 cgroup.max.depth (since Linux 4.14)
643 This file defines a limit on the depth of nesting of descendant
644 cgroups. A value of 0 in this file means that no descendant
645 cgroups can be created. An attempt to create a descendant whose
646 nesting level exceeds the limit fails (mkdir(2) fails with the
647 error EAGAIN).
648
649 Writing the string "max" to this file means that no limit is im‐
650 posed. The default value in this file is "max".
651
652 cgroup.max.descendants (since Linux 4.14)
653 This file defines a limit on the number of live descendant
654 cgroups that this cgroup may have. An attempt to create more
655 descendants than allowed by the limit fails (mkdir(2) fails with
656 the error EAGAIN).
657
658 Writing the string "max" to this file means that no limit is im‐
659 posed. The default value in this file is "max".
660
662 In the context of cgroups, delegation means passing management of some
663 subtree of the cgroup hierarchy to a nonprivileged user. Cgroups v1
664 provides support for delegation based on file permissions in the cgroup
665 hierarchy but with less strict containment rules than v2 (as noted be‐
666 low). Cgroups v2 supports delegation with containment by explicit de‐
667 sign. The focus of the discussion in this section is on delegation in
668 cgroups v2, with some differences for cgroups v1 noted along the way.
669
670 Some terminology is required in order to describe delegation. A dele‐
671 gater is a privileged user (i.e., root) who owns a parent cgroup. A
672 delegatee is a nonprivileged user who will be granted the permissions
673 needed to manage some subhierarchy under that parent cgroup, known as
674 the delegated subtree.
675
676 To perform delegation, the delegater makes certain directories and
677 files writable by the delegatee, typically by changing the ownership of
678 the objects to be the user ID of the delegatee. Assuming that we want
679 to delegate the hierarchy rooted at (say) /dlgt_grp and that there are
680 not yet any child cgroups under that cgroup, the ownership of the fol‐
681 lowing is changed to the user ID of the delegatee:
682
683 /dlgt_grp
684 Changing the ownership of the root of the subtree means that any
685 new cgroups created under the subtree (and the files they con‐
686 tain) will also be owned by the delegatee.
687
688 /dlgt_grp/cgroup.procs
689 Changing the ownership of this file means that the delegatee can
690 move processes into the root of the delegated subtree.
691
692 /dlgt_grp/cgroup.subtree_control (cgroups v2 only)
693 Changing the ownership of this file means that the delegatee can
694 enable controllers (that are present in /dlgt_grp/cgroup.con‐
695 trollers) in order to further redistribute resources at lower
696 levels in the subtree. (As an alternative to changing the own‐
697 ership of this file, the delegater might instead add selected
698 controllers to this file.)
699
700 /dlgt_grp/cgroup.threads (cgroups v2 only)
701 Changing the ownership of this file is necessary if a threaded
702 subtree is being delegated (see the description of "thread
703 mode", below). This permits the delegatee to write thread IDs
704 to the file. (The ownership of this file can also be changed
705 when delegating a domain subtree, but currently this serves no
706 purpose, since, as described below, it is not possible to move a
707 thread between domain cgroups by writing its thread ID to the
708 cgroup.threads file.)
709
710 In cgroups v1, the corresponding file that should instead be
711 delegated is the tasks file.
712
713 The delegater should not change the ownership of any of the controller
714 interfaces files (e.g., pids.max, memory.high) in dlgt_grp. Those
715 files are used from the next level above the delegated subtree in order
716 to distribute resources into the subtree, and the delegatee should not
717 have permission to change the resources that are distributed into the
718 delegated subtree.
719
720 See also the discussion of the /sys/kernel/cgroup/delegate file in
721 NOTES for information about further delegatable files in cgroups v2.
722
723 After the aforementioned steps have been performed, the delegatee can
724 create child cgroups within the delegated subtree (the cgroup subdirec‐
725 tories and the files they contain will be owned by the delegatee) and
726 move processes between cgroups in the subtree. If some controllers are
727 present in dlgt_grp/cgroup.subtree_control, or the ownership of that
728 file was passed to the delegatee, the delegatee can also control the
729 further redistribution of the corresponding resources into the dele‐
730 gated subtree.
731
732 Cgroups v2 delegation: nsdelegate and cgroup namespaces
733 Starting with Linux 4.13, there is a second way to perform cgroup dele‐
734 gation in the cgroups v2 hierarchy. This is done by mounting or re‐
735 mounting the cgroup v2 filesystem with the nsdelegate mount option.
736 For example, if the cgroup v2 filesystem has already been mounted, we
737 can remount it with the nsdelegate option as follows:
738
739 mount -t cgroup2 -o remount,nsdelegate \
740 none /sys/fs/cgroup/unified
741
742 The effect of this mount option is to cause cgroup namespaces to auto‐
743 matically become delegation boundaries. More specifically, the follow‐
744 ing restrictions apply for processes inside the cgroup namespace:
745
746 * Writes to controller interface files in the root directory of the
747 namespace will fail with the error EPERM. Processes inside the
748 cgroup namespace can still write to delegatable files in the root
749 directory of the cgroup namespace such as cgroup.procs and
750 cgroup.subtree_control, and can create subhierarchy underneath the
751 root directory.
752
753 * Attempts to migrate processes across the namespace boundary are de‐
754 nied (with the error ENOENT). Processes inside the cgroup namespace
755 can still (subject to the containment rules described below) move
756 processes between cgroups within the subhierarchy under the name‐
757 space root.
758
759 The ability to define cgroup namespaces as delegation boundaries makes
760 cgroup namespaces more useful. To understand why, suppose that we al‐
761 ready have one cgroup hierarchy that has been delegated to a nonprivi‐
762 leged user, cecilia, using the older delegation technique described
763 above. Suppose further that cecilia wanted to further delegate a sub‐
764 hierarchy under the existing delegated hierarchy. (For example, the
765 delegated hierarchy might be associated with an unprivileged container
766 run by cecilia.) Even if a cgroup namespace was employed, because both
767 hierarchies are owned by the unprivileged user cecilia, the following
768 illegitimate actions could be performed:
769
770 * A process in the inferior hierarchy could change the resource con‐
771 troller settings in the root directory of that hierarchy. (These
772 resource controller settings are intended to allow control to be ex‐
773 ercised from the parent cgroup; a process inside the child cgroup
774 should not be allowed to modify them.)
775
776 * A process inside the inferior hierarchy could move processes into
777 and out of the inferior hierarchy if the cgroups in the superior hi‐
778 erarchy were somehow visible.
779
780 Employing the nsdelegate mount option prevents both of these possibili‐
781 ties.
782
783 The nsdelegate mount option only has an effect when performed in the
784 initial mount namespace; in other mount namespaces, the option is
785 silently ignored.
786
787 Note: On some systems, systemd(1) automatically mounts the cgroup v2
788 filesystem. In order to experiment with the nsdelegate operation, it
789 may be useful to boot the kernel with the following command-line op‐
790 tions:
791
792 cgroup_no_v1=all systemd.legacy_systemd_cgroup_controller
793
794 These options cause the kernel to boot with the cgroups v1 controllers
795 disabled (meaning that the controllers are available in the v2 hierar‐
796 chy), and tells systemd(1) not to mount and use the cgroup v2 hierar‐
797 chy, so that the v2 hierarchy can be manually mounted with the desired
798 options after boot-up.
799
800 Cgroup delegation containment rules
801 Some delegation containment rules ensure that the delegatee can move
802 processes between cgroups within the delegated subtree, but can't move
803 processes from outside the delegated subtree into the subtree or vice
804 versa. A nonprivileged process (i.e., the delegatee) can write the PID
805 of a "target" process into a cgroup.procs file only if all of the fol‐
806 lowing are true:
807
808 * The writer has write permission on the cgroup.procs file in the des‐
809 tination cgroup.
810
811 * The writer has write permission on the cgroup.procs file in the
812 nearest common ancestor of the source and destination cgroups. Note
813 that in some cases, the nearest common ancestor may be the source or
814 destination cgroup itself. This requirement is not enforced for
815 cgroups v1 hierarchies, with the consequence that containment in v1
816 is less strict than in v2. (For example, in cgroups v1 the user
817 that owns two distinct delegated subhierarchies can move a process
818 between the hierarchies.)
819
820 * If the cgroup v2 filesystem was mounted with the nsdelegate option,
821 the writer must be able to see the source and destination cgroups
822 from its cgroup namespace.
823
824 * In cgroups v1: the effective UID of the writer (i.e., the delegatee)
825 matches the real user ID or the saved set-user-ID of the target
826 process. Before Linux 4.11, this requirement also applied in
827 cgroups v2 (This was a historical requirement inherited from cgroups
828 v1 that was later deemed unnecessary, since the other rules suffice
829 for containment in cgroups v2.)
830
831 Note: one consequence of these delegation containment rules is that the
832 unprivileged delegatee can't place the first process into the delegated
833 subtree; instead, the delegater must place the first process (a process
834 owned by the delegatee) into the delegated subtree.
835
837 Among the restrictions imposed by cgroups v2 that were not present in
838 cgroups v1 are the following:
839
840 * No thread-granularity control: all of the threads of a process must
841 be in the same cgroup.
842
843 * No internal processes: a cgroup can't both have member processes and
844 exercise controllers on child cgroups.
845
846 Both of these restrictions were added because the lack of these re‐
847 strictions had caused problems in cgroups v1. In particular, the
848 cgroups v1 ability to allow thread-level granularity for cgroup member‐
849 ship made no sense for some controllers. (A notable example was the
850 memory controller: since threads share an address space, it made no
851 sense to split threads across different memory cgroups.)
852
853 Notwithstanding the initial design decision in cgroups v2, there were
854 use cases for certain controllers, notably the cpu controller, for
855 which thread-level granularity of control was meaningful and useful.
856 To accommodate such use cases, Linux 4.14 added thread mode for cgroups
857 v2.
858
859 Thread mode allows the following:
860
861 * The creation of threaded subtrees in which the threads of a process
862 may be spread across cgroups inside the tree. (A threaded subtree
863 may contain multiple multithreaded processes.)
864
865 * The concept of threaded controllers, which can distribute resources
866 across the cgroups in a threaded subtree.
867
868 * A relaxation of the "no internal processes rule", so that, within a
869 threaded subtree, a cgroup can both contain member threads and exer‐
870 cise resource control over child cgroups.
871
872 With the addition of thread mode, each nonroot cgroup now contains a
873 new file, cgroup.type, that exposes, and in some circumstances can be
874 used to change, the "type" of a cgroup. This file contains one of the
875 following type values:
876
877 domain This is a normal v2 cgroup that provides process-granularity
878 control. If a process is a member of this cgroup, then all
879 threads of the process are (by definition) in the same cgroup.
880 This is the default cgroup type, and provides the same behavior
881 that was provided for cgroups in the initial cgroups v2 imple‐
882 mentation.
883
884 threaded
885 This cgroup is a member of a threaded subtree. Threads can be
886 added to this cgroup, and controllers can be enabled for the
887 cgroup.
888
889 domain threaded
890 This is a domain cgroup that serves as the root of a threaded
891 subtree. This cgroup type is also known as "threaded root".
892
893 domain invalid
894 This is a cgroup inside a threaded subtree that is in an "in‐
895 valid" state. Processes can't be added to the cgroup, and con‐
896 trollers can't be enabled for the cgroup. The only thing that
897 can be done with this cgroup (other than deleting it) is to con‐
898 vert it to a threaded cgroup by writing the string "threaded" to
899 the cgroup.type file.
900
901 The rationale for the existence of this "interim" type during
902 the creation of a threaded subtree (rather than the kernel sim‐
903 ply immediately converting all cgroups under the threaded root
904 to the type threaded) is to allow for possible future extensions
905 to the thread mode model
906
907 Threaded versus domain controllers
908 With the addition of threads mode, cgroups v2 now distinguishes two
909 types of resource controllers:
910
911 * Threaded controllers: these controllers support thread-granularity
912 for resource control and can be enabled inside threaded subtrees,
913 with the result that the corresponding controller-interface files
914 appear inside the cgroups in the threaded subtree. As at Linux
915 4.19, the following controllers are threaded: cpu, perf_event, and
916 pids.
917
918 * Domain controllers: these controllers support only process granular‐
919 ity for resource control. From the perspective of a domain con‐
920 troller, all threads of a process are always in the same cgroup.
921 Domain controllers can't be enabled inside a threaded subtree.
922
923 Creating a threaded subtree
924 There are two pathways that lead to the creation of a threaded subtree.
925 The first pathway proceeds as follows:
926
927 1. We write the string "threaded" to the cgroup.type file of a cgroup
928 y/z that currently has the type domain. This has the following ef‐
929 fects:
930
931 * The type of the cgroup y/z becomes threaded.
932
933 * The type of the parent cgroup, y, becomes domain threaded. The
934 parent cgroup is the root of a threaded subtree (also known as
935 the "threaded root").
936
937 * All other cgroups under y that were not already of type threaded
938 (because they were inside already existing threaded subtrees un‐
939 der the new threaded root) are converted to type domain invalid.
940 Any subsequently created cgroups under y will also have the type
941 domain invalid.
942
943 2. We write the string "threaded" to each of the domain invalid cgroups
944 under y, in order to convert them to the type threaded. As a conse‐
945 quence of this step, all threads under the threaded root now have
946 the type threaded and the threaded subtree is now fully usable. The
947 requirement to write "threaded" to each of these cgroups is somewhat
948 cumbersome, but allows for possible future extensions to the thread-
949 mode model.
950
951 The second way of creating a threaded subtree is as follows:
952
953 1. In an existing cgroup, z, that currently has the type domain, we (1)
954 enable one or more threaded controllers and (2) make a process a
955 member of z. (These two steps can be done in either order.) This
956 has the following consequences:
957
958 * The type of z becomes domain threaded.
959
960 * All of the descendant cgroups of x that were not already of type
961 threaded are converted to type domain invalid.
962
963 2. As before, we make the threaded subtree usable by writing the string
964 "threaded" to each of the domain invalid cgroups under y, in order
965 to convert them to the type threaded.
966
967 One of the consequences of the above pathways to creating a threaded
968 subtree is that the threaded root cgroup can be a parent only to
969 threaded (and domain invalid) cgroups. The threaded root cgroup can't
970 be a parent of a domain cgroups, and a threaded cgroup can't have a
971 sibling that is a domain cgroup.
972
973 Using a threaded subtree
974 Within a threaded subtree, threaded controllers can be enabled in each
975 subgroup whose type has been changed to threaded; upon doing so, the
976 corresponding controller interface files appear in the children of that
977 cgroup.
978
979 A process can be moved into a threaded subtree by writing its PID to
980 the cgroup.procs file in one of the cgroups inside the tree. This has
981 the effect of making all of the threads in the process members of the
982 corresponding cgroup and makes the process a member of the threaded
983 subtree. The threads of the process can then be spread across the
984 threaded subtree by writing their thread IDs (see gettid(2)) to the
985 cgroup.threads files in different cgroups inside the subtree. The
986 threads of a process must all reside in the same threaded subtree.
987
988 As with writing to cgroup.procs, some containment rules apply when
989 writing to the cgroup.threads file:
990
991 * The writer must have write permission on the cgroup.threads file in
992 the destination cgroup.
993
994 * The writer must have write permission on the cgroup.procs file in
995 the common ancestor of the source and destination cgroups. (In some
996 cases, the common ancestor may be the source or destination cgroup
997 itself.)
998
999 * The source and destination cgroups must be in the same threaded sub‐
1000 tree. (Outside a threaded subtree, an attempt to move a thread by
1001 writing its thread ID to the cgroup.threads file in a different do‐
1002 main cgroup fails with the error EOPNOTSUPP.)
1003
1004 The cgroup.threads file is present in each cgroup (including domain
1005 cgroups) and can be read in order to discover the set of threads that
1006 is present in the cgroup. The set of thread IDs obtained when reading
1007 this file is not guaranteed to be ordered or free of duplicates.
1008
1009 The cgroup.procs file in the threaded root shows the PIDs of all pro‐
1010 cesses that are members of the threaded subtree. The cgroup.procs
1011 files in the other cgroups in the subtree are not readable.
1012
1013 Domain controllers can't be enabled in a threaded subtree; no con‐
1014 troller-interface files appear inside the cgroups underneath the
1015 threaded root. From the point of view of a domain controller, threaded
1016 subtrees are invisible: a multithreaded process inside a threaded sub‐
1017 tree appears to a domain controller as a process that resides in the
1018 threaded root cgroup.
1019
1020 Within a threaded subtree, the "no internal processes" rule does not
1021 apply: a cgroup can both contain member processes (or thread) and exer‐
1022 cise controllers on child cgroups.
1023
1024 Rules for writing to cgroup.type and creating threaded subtrees
1025 A number of rules apply when writing to the cgroup.type file:
1026
1027 * Only the string "threaded" may be written. In other words, the only
1028 explicit transition that is possible is to convert a domain cgroup
1029 to type threaded.
1030
1031 * The effect of writing "threaded" depends on the current value in
1032 cgroup.type, as follows:
1033
1034 • domain or domain threaded: start the creation of a threaded sub‐
1035 tree (whose root is the parent of this cgroup) via the first of
1036 the pathways described above;
1037
1038 • domain invalid: convert this cgroup (which is inside a threaded
1039 subtree) to a usable (i.e., threaded) state;
1040
1041 • threaded: no effect (a "no-op").
1042
1043 * We can't write to a cgroup.type file if the parent's type is domain
1044 invalid. In other words, the cgroups of a threaded subtree must be
1045 converted to the threaded state in a top-down manner.
1046
1047 There are also some constraints that must be satisfied in order to cre‐
1048 ate a threaded subtree rooted at the cgroup x:
1049
1050 * There can be no member processes in the descendant cgroups of x.
1051 (The cgroup x can itself have member processes.)
1052
1053 * No domain controllers may be enabled in x's cgroup.subtree_control
1054 file.
1055
1056 If any of the above constraints is violated, then an attempt to write
1057 "threaded" to a cgroup.type file fails with the error ENOTSUP.
1058
1059 The "domain threaded" cgroup type
1060 According to the pathways described above, the type of a cgroup can
1061 change to domain threaded in either of the following cases:
1062
1063 * The string "threaded" is written to a child cgroup.
1064
1065 * A threaded controller is enabled inside the cgroup and a process is
1066 made a member of the cgroup.
1067
1068 A domain threaded cgroup, x, can revert to the type domain if the above
1069 conditions no longer hold true—that is, if all threaded child cgroups
1070 of x are removed and either x no longer has threaded controllers en‐
1071 abled or no longer has member processes.
1072
1073 When a domain threaded cgroup x reverts to the type domain:
1074
1075 * All domain invalid descendants of x that are not in lower-level
1076 threaded subtrees revert to the type domain.
1077
1078 * The root cgroups in any lower-level threaded subtrees revert to the
1079 type domain threaded.
1080
1081 Exceptions for the root cgroup
1082 The root cgroup of the v2 hierarchy is treated exceptionally: it can be
1083 the parent of both domain and threaded cgroups. If the string
1084 "threaded" is written to the cgroup.type file of one of the children of
1085 the root cgroup, then
1086
1087 * The type of that cgroup becomes threaded.
1088
1089 * The type of any descendants of that cgroup that are not part of
1090 lower-level threaded subtrees changes to domain invalid.
1091
1092 Note that in this case, there is no cgroup whose type becomes domain
1093 threaded. (Notionally, the root cgroup can be considered as the
1094 threaded root for the cgroup whose type was changed to threaded.)
1095
1096 The aim of this exceptional treatment for the root cgroup is to allow a
1097 threaded cgroup that employs the cpu controller to be placed as high as
1098 possible in the hierarchy, so as to minimize the (small) cost of
1099 traversing the cgroup hierarchy.
1100
1101 The cgroups v2 "cpu" controller and realtime threads
1102 As at Linux 4.19, the cgroups v2 cpu controller does not support con‐
1103 trol of realtime threads (specifically threads scheduled under any of
1104 the policies SCHED_FIFO, SCHED_RR, described SCHED_DEADLINE; see
1105 sched(7)). Therefore, the cpu controller can be enabled in the root
1106 cgroup only if all realtime threads are in the root cgroup. (If there
1107 are realtime threads in nonroot cgroups, then a write(2) of the string
1108 "+cpu" to the cgroup.subtree_control file fails with the error EINVAL.)
1109
1110 On some systems, systemd(1) places certain realtime threads in nonroot
1111 cgroups in the v2 hierarchy. On such systems, these threads must first
1112 be moved to the root cgroup before the cpu controller can be enabled.
1113
1115 The following errors can occur for mount(2):
1116
1117 EBUSY An attempt to mount a cgroup version 1 filesystem specified nei‐
1118 ther the name= option (to mount a named hierarchy) nor a con‐
1119 troller name (or all).
1120
1122 A child process created via fork(2) inherits its parent's cgroup mem‐
1123 berships. A process's cgroup memberships are preserved across ex‐
1124 ecve(2).
1125
1126 The clone3(2) CLONE_INTO_CGROUP flag can be used to create a child
1127 process that begins its life in a different version 2 cgroup from the
1128 parent process.
1129
1130 /proc files
1131 /proc/cgroups (since Linux 2.6.24)
1132 This file contains information about the controllers that are
1133 compiled into the kernel. An example of the contents of this
1134 file (reformatted for readability) is the following:
1135
1136 #subsys_name hierarchy num_cgroups enabled
1137 cpuset 4 1 1
1138 cpu 8 1 1
1139 cpuacct 8 1 1
1140 blkio 6 1 1
1141 memory 3 1 1
1142 devices 10 84 1
1143 freezer 7 1 1
1144 net_cls 9 1 1
1145 perf_event 5 1 1
1146 net_prio 9 1 1
1147 hugetlb 0 1 0
1148 pids 2 1 1
1149
1150 The fields in this file are, from left to right:
1151
1152 1. The name of the controller.
1153
1154 2. The unique ID of the cgroup hierarchy on which this con‐
1155 troller is mounted. If multiple cgroups v1 controllers are
1156 bound to the same hierarchy, then each will show the same hi‐
1157 erarchy ID in this field. The value in this field will be 0
1158 if:
1159
1160 a) the controller is not mounted on a cgroups v1 hierarchy;
1161
1162 b) the controller is bound to the cgroups v2 single unified
1163 hierarchy; or
1164
1165 c) the controller is disabled (see below).
1166
1167 3. The number of control groups in this hierarchy using this
1168 controller.
1169
1170 4. This field contains the value 1 if this controller is en‐
1171 abled, or 0 if it has been disabled (via the cgroup_disable
1172 kernel command-line boot parameter).
1173
1174 /proc/[pid]/cgroup (since Linux 2.6.24)
1175 This file describes control groups to which the process with the
1176 corresponding PID belongs. The displayed information differs
1177 for cgroups version 1 and version 2 hierarchies.
1178
1179 For each cgroup hierarchy of which the process is a member,
1180 there is one entry containing three colon-separated fields:
1181
1182 hierarchy-ID:controller-list:cgroup-path
1183
1184 For example:
1185
1186 5:cpuacct,cpu,cpuset:/daemons
1187
1188 The colon-separated fields are, from left to right:
1189
1190 1. For cgroups version 1 hierarchies, this field contains a
1191 unique hierarchy ID number that can be matched to a hierarchy
1192 ID in /proc/cgroups. For the cgroups version 2 hierarchy,
1193 this field contains the value 0.
1194
1195 2. For cgroups version 1 hierarchies, this field contains a
1196 comma-separated list of the controllers bound to the hierar‐
1197 chy. For the cgroups version 2 hierarchy, this field is
1198 empty.
1199
1200 3. This field contains the pathname of the control group in the
1201 hierarchy to which the process belongs. This pathname is
1202 relative to the mount point of the hierarchy.
1203
1204 /sys/kernel/cgroup files
1205 /sys/kernel/cgroup/delegate (since Linux 4.15)
1206 This file exports a list of the cgroups v2 files (one per line)
1207 that are delegatable (i.e., whose ownership should be changed to
1208 the user ID of the delegatee). In the future, the set of dele‐
1209 gatable files may change or grow, and this file provides a way
1210 for the kernel to inform user-space applications of which files
1211 must be delegated. As at Linux 4.15, one sees the following
1212 when inspecting this file:
1213
1214 $ cat /sys/kernel/cgroup/delegate
1215 cgroup.procs
1216 cgroup.subtree_control
1217 cgroup.threads
1218
1219 /sys/kernel/cgroup/features (since Linux 4.15)
1220 Over time, the set of cgroups v2 features that are provided by
1221 the kernel may change or grow, or some features may not be en‐
1222 abled by default. This file provides a way for user-space ap‐
1223 plications to discover what features the running kernel supports
1224 and has enabled. Features are listed one per line:
1225
1226 $ cat /sys/kernel/cgroup/features
1227 nsdelegate
1228 memory_localevents
1229
1230 The entries that can appear in this file are:
1231
1232 memory_localevents (since Linux 5.2)
1233 The kernel supports the memory_localevents mount option.
1234
1235 nsdelegate (since Linux 4.15)
1236 The kernel supports the nsdelegate mount option.
1237
1239 prlimit(1), systemd(1), systemd-cgls(1), systemd-cgtop(1), clone(2),
1240 ioprio_set(2), perf_event_open(2), setrlimit(2), cgroup_namespaces(7),
1241 cpuset(7), namespaces(7), sched(7), user_namespaces(7)
1242
1243 The kernel source file Documentation/admin-guide/cgroup-v2.rst.
1244
1246 This page is part of release 5.13 of the Linux man-pages project. A
1247 description of the project, information about reporting bugs, and the
1248 latest version of this page, can be found at
1249 https://www.kernel.org/doc/man-pages/.
1250
1251
1252
1253Linux 2021-08-27 CGROUPS(7)