xref: /linux/Documentation/admin-guide/cgroup-v2.rst (revision b0f84a84fff180718995b1269da2988e5b28be42)
1================
2Control Group v2
3================
4
5:Date: October, 2015
6:Author: Tejun Heo <tj@kernel.org>
7
8This is the authoritative documentation on the design, interface and
9conventions of cgroup v2.  It describes all userland-visible aspects
10of cgroup including core and specific controller behaviors.  All
11future changes must be reflected in this document.  Documentation for
12v1 is available under Documentation/cgroup-v1/.
13
14.. CONTENTS
15
16   1. Introduction
17     1-1. Terminology
18     1-2. What is cgroup?
19   2. Basic Operations
20     2-1. Mounting
21     2-2. Organizing Processes and Threads
22       2-2-1. Processes
23       2-2-2. Threads
24     2-3. [Un]populated Notification
25     2-4. Controlling Controllers
26       2-4-1. Enabling and Disabling
27       2-4-2. Top-down Constraint
28       2-4-3. No Internal Process Constraint
29     2-5. Delegation
30       2-5-1. Model of Delegation
31       2-5-2. Delegation Containment
32     2-6. Guidelines
33       2-6-1. Organize Once and Control
34       2-6-2. Avoid Name Collisions
35   3. Resource Distribution Models
36     3-1. Weights
37     3-2. Limits
38     3-3. Protections
39     3-4. Allocations
40   4. Interface Files
41     4-1. Format
42     4-2. Conventions
43     4-3. Core Interface Files
44   5. Controllers
45     5-1. CPU
46       5-1-1. CPU Interface Files
47     5-2. Memory
48       5-2-1. Memory Interface Files
49       5-2-2. Usage Guidelines
50       5-2-3. Memory Ownership
51     5-3. IO
52       5-3-1. IO Interface Files
53       5-3-2. Writeback
54       5-3-3. IO Latency
55         5-3-3-1. How IO Latency Throttling Works
56         5-3-3-2. IO Latency Interface Files
57     5-4. PID
58       5-4-1. PID Interface Files
59     5-5. Cpuset
60       5.5-1. Cpuset Interface Files
61     5-6. Device
62     5-7. RDMA
63       5-7-1. RDMA Interface Files
64     5-8. Misc
65       5-8-1. perf_event
66     5-N. Non-normative information
67       5-N-1. CPU controller root cgroup process behaviour
68       5-N-2. IO controller root cgroup process behaviour
69   6. Namespace
70     6-1. Basics
71     6-2. The Root and Views
72     6-3. Migration and setns(2)
73     6-4. Interaction with Other Namespaces
74   P. Information on Kernel Programming
75     P-1. Filesystem Support for Writeback
76   D. Deprecated v1 Core Features
77   R. Issues with v1 and Rationales for v2
78     R-1. Multiple Hierarchies
79     R-2. Thread Granularity
80     R-3. Competition Between Inner Nodes and Threads
81     R-4. Other Interface Issues
82     R-5. Controller Issues and Remedies
83       R-5-1. Memory
84
85
86Introduction
87============
88
89Terminology
90-----------
91
92"cgroup" stands for "control group" and is never capitalized.  The
93singular form is used to designate the whole feature and also as a
94qualifier as in "cgroup controllers".  When explicitly referring to
95multiple individual control groups, the plural form "cgroups" is used.
96
97
98What is cgroup?
99---------------
100
101cgroup is a mechanism to organize processes hierarchically and
102distribute system resources along the hierarchy in a controlled and
103configurable manner.
104
105cgroup is largely composed of two parts - the core and controllers.
106cgroup core is primarily responsible for hierarchically organizing
107processes.  A cgroup controller is usually responsible for
108distributing a specific type of system resource along the hierarchy
109although there are utility controllers which serve purposes other than
110resource distribution.
111
112cgroups form a tree structure and every process in the system belongs
113to one and only one cgroup.  All threads of a process belong to the
114same cgroup.  On creation, all processes are put in the cgroup that
115the parent process belongs to at the time.  A process can be migrated
116to another cgroup.  Migration of a process doesn't affect already
117existing descendant processes.
118
119Following certain structural constraints, controllers may be enabled or
120disabled selectively on a cgroup.  All controller behaviors are
121hierarchical - if a controller is enabled on a cgroup, it affects all
122processes which belong to the cgroups consisting the inclusive
123sub-hierarchy of the cgroup.  When a controller is enabled on a nested
124cgroup, it always restricts the resource distribution further.  The
125restrictions set closer to the root in the hierarchy can not be
126overridden from further away.
127
128
129Basic Operations
130================
131
132Mounting
133--------
134
135Unlike v1, cgroup v2 has only single hierarchy.  The cgroup v2
136hierarchy can be mounted with the following mount command::
137
138  # mount -t cgroup2 none $MOUNT_POINT
139
140cgroup2 filesystem has the magic number 0x63677270 ("cgrp").  All
141controllers which support v2 and are not bound to a v1 hierarchy are
142automatically bound to the v2 hierarchy and show up at the root.
143Controllers which are not in active use in the v2 hierarchy can be
144bound to other hierarchies.  This allows mixing v2 hierarchy with the
145legacy v1 multiple hierarchies in a fully backward compatible way.
146
147A controller can be moved across hierarchies only after the controller
148is no longer referenced in its current hierarchy.  Because per-cgroup
149controller states are destroyed asynchronously and controllers may
150have lingering references, a controller may not show up immediately on
151the v2 hierarchy after the final umount of the previous hierarchy.
152Similarly, a controller should be fully disabled to be moved out of
153the unified hierarchy and it may take some time for the disabled
154controller to become available for other hierarchies; furthermore, due
155to inter-controller dependencies, other controllers may need to be
156disabled too.
157
158While useful for development and manual configurations, moving
159controllers dynamically between the v2 and other hierarchies is
160strongly discouraged for production use.  It is recommended to decide
161the hierarchies and controller associations before starting using the
162controllers after system boot.
163
164During transition to v2, system management software might still
165automount the v1 cgroup filesystem and so hijack all controllers
166during boot, before manual intervention is possible. To make testing
167and experimenting easier, the kernel parameter cgroup_no_v1= allows
168disabling controllers in v1 and make them always available in v2.
169
170cgroup v2 currently supports the following mount options.
171
172  nsdelegate
173
174	Consider cgroup namespaces as delegation boundaries.  This
175	option is system wide and can only be set on mount or modified
176	through remount from the init namespace.  The mount option is
177	ignored on non-init namespace mounts.  Please refer to the
178	Delegation section for details.
179
180  memory_localevents
181
182        Only populate memory.events with data for the current cgroup,
183        and not any subtrees. This is legacy behaviour, the default
184        behaviour without this option is to include subtree counts.
185        This option is system wide and can only be set on mount or
186        modified through remount from the init namespace. The mount
187        option is ignored on non-init namespace mounts.
188
189
190Organizing Processes and Threads
191--------------------------------
192
193Processes
194~~~~~~~~~
195
196Initially, only the root cgroup exists to which all processes belong.
197A child cgroup can be created by creating a sub-directory::
198
199  # mkdir $CGROUP_NAME
200
201A given cgroup may have multiple child cgroups forming a tree
202structure.  Each cgroup has a read-writable interface file
203"cgroup.procs".  When read, it lists the PIDs of all processes which
204belong to the cgroup one-per-line.  The PIDs are not ordered and the
205same PID may show up more than once if the process got moved to
206another cgroup and then back or the PID got recycled while reading.
207
208A process can be migrated into a cgroup by writing its PID to the
209target cgroup's "cgroup.procs" file.  Only one process can be migrated
210on a single write(2) call.  If a process is composed of multiple
211threads, writing the PID of any thread migrates all threads of the
212process.
213
214When a process forks a child process, the new process is born into the
215cgroup that the forking process belongs to at the time of the
216operation.  After exit, a process stays associated with the cgroup
217that it belonged to at the time of exit until it's reaped; however, a
218zombie process does not appear in "cgroup.procs" and thus can't be
219moved to another cgroup.
220
221A cgroup which doesn't have any children or live processes can be
222destroyed by removing the directory.  Note that a cgroup which doesn't
223have any children and is associated only with zombie processes is
224considered empty and can be removed::
225
226  # rmdir $CGROUP_NAME
227
228"/proc/$PID/cgroup" lists a process's cgroup membership.  If legacy
229cgroup is in use in the system, this file may contain multiple lines,
230one for each hierarchy.  The entry for cgroup v2 is always in the
231format "0::$PATH"::
232
233  # cat /proc/842/cgroup
234  ...
235  0::/test-cgroup/test-cgroup-nested
236
237If the process becomes a zombie and the cgroup it was associated with
238is removed subsequently, " (deleted)" is appended to the path::
239
240  # cat /proc/842/cgroup
241  ...
242  0::/test-cgroup/test-cgroup-nested (deleted)
243
244
245Threads
246~~~~~~~
247
248cgroup v2 supports thread granularity for a subset of controllers to
249support use cases requiring hierarchical resource distribution across
250the threads of a group of processes.  By default, all threads of a
251process belong to the same cgroup, which also serves as the resource
252domain to host resource consumptions which are not specific to a
253process or thread.  The thread mode allows threads to be spread across
254a subtree while still maintaining the common resource domain for them.
255
256Controllers which support thread mode are called threaded controllers.
257The ones which don't are called domain controllers.
258
259Marking a cgroup threaded makes it join the resource domain of its
260parent as a threaded cgroup.  The parent may be another threaded
261cgroup whose resource domain is further up in the hierarchy.  The root
262of a threaded subtree, that is, the nearest ancestor which is not
263threaded, is called threaded domain or thread root interchangeably and
264serves as the resource domain for the entire subtree.
265
266Inside a threaded subtree, threads of a process can be put in
267different cgroups and are not subject to the no internal process
268constraint - threaded controllers can be enabled on non-leaf cgroups
269whether they have threads in them or not.
270
271As the threaded domain cgroup hosts all the domain resource
272consumptions of the subtree, it is considered to have internal
273resource consumptions whether there are processes in it or not and
274can't have populated child cgroups which aren't threaded.  Because the
275root cgroup is not subject to no internal process constraint, it can
276serve both as a threaded domain and a parent to domain cgroups.
277
278The current operation mode or type of the cgroup is shown in the
279"cgroup.type" file which indicates whether the cgroup is a normal
280domain, a domain which is serving as the domain of a threaded subtree,
281or a threaded cgroup.
282
283On creation, a cgroup is always a domain cgroup and can be made
284threaded by writing "threaded" to the "cgroup.type" file.  The
285operation is single direction::
286
287  # echo threaded > cgroup.type
288
289Once threaded, the cgroup can't be made a domain again.  To enable the
290thread mode, the following conditions must be met.
291
292- As the cgroup will join the parent's resource domain.  The parent
293  must either be a valid (threaded) domain or a threaded cgroup.
294
295- When the parent is an unthreaded domain, it must not have any domain
296  controllers enabled or populated domain children.  The root is
297  exempt from this requirement.
298
299Topology-wise, a cgroup can be in an invalid state.  Please consider
300the following topology::
301
302  A (threaded domain) - B (threaded) - C (domain, just created)
303
304C is created as a domain but isn't connected to a parent which can
305host child domains.  C can't be used until it is turned into a
306threaded cgroup.  "cgroup.type" file will report "domain (invalid)" in
307these cases.  Operations which fail due to invalid topology use
308EOPNOTSUPP as the errno.
309
310A domain cgroup is turned into a threaded domain when one of its child
311cgroup becomes threaded or threaded controllers are enabled in the
312"cgroup.subtree_control" file while there are processes in the cgroup.
313A threaded domain reverts to a normal domain when the conditions
314clear.
315
316When read, "cgroup.threads" contains the list of the thread IDs of all
317threads in the cgroup.  Except that the operations are per-thread
318instead of per-process, "cgroup.threads" has the same format and
319behaves the same way as "cgroup.procs".  While "cgroup.threads" can be
320written to in any cgroup, as it can only move threads inside the same
321threaded domain, its operations are confined inside each threaded
322subtree.
323
324The threaded domain cgroup serves as the resource domain for the whole
325subtree, and, while the threads can be scattered across the subtree,
326all the processes are considered to be in the threaded domain cgroup.
327"cgroup.procs" in a threaded domain cgroup contains the PIDs of all
328processes in the subtree and is not readable in the subtree proper.
329However, "cgroup.procs" can be written to from anywhere in the subtree
330to migrate all threads of the matching process to the cgroup.
331
332Only threaded controllers can be enabled in a threaded subtree.  When
333a threaded controller is enabled inside a threaded subtree, it only
334accounts for and controls resource consumptions associated with the
335threads in the cgroup and its descendants.  All consumptions which
336aren't tied to a specific thread belong to the threaded domain cgroup.
337
338Because a threaded subtree is exempt from no internal process
339constraint, a threaded controller must be able to handle competition
340between threads in a non-leaf cgroup and its child cgroups.  Each
341threaded controller defines how such competitions are handled.
342
343
344[Un]populated Notification
345--------------------------
346
347Each non-root cgroup has a "cgroup.events" file which contains
348"populated" field indicating whether the cgroup's sub-hierarchy has
349live processes in it.  Its value is 0 if there is no live process in
350the cgroup and its descendants; otherwise, 1.  poll and [id]notify
351events are triggered when the value changes.  This can be used, for
352example, to start a clean-up operation after all processes of a given
353sub-hierarchy have exited.  The populated state updates and
354notifications are recursive.  Consider the following sub-hierarchy
355where the numbers in the parentheses represent the numbers of processes
356in each cgroup::
357
358  A(4) - B(0) - C(1)
359              \ D(0)
360
361A, B and C's "populated" fields would be 1 while D's 0.  After the one
362process in C exits, B and C's "populated" fields would flip to "0" and
363file modified events will be generated on the "cgroup.events" files of
364both cgroups.
365
366
367Controlling Controllers
368-----------------------
369
370Enabling and Disabling
371~~~~~~~~~~~~~~~~~~~~~~
372
373Each cgroup has a "cgroup.controllers" file which lists all
374controllers available for the cgroup to enable::
375
376  # cat cgroup.controllers
377  cpu io memory
378
379No controller is enabled by default.  Controllers can be enabled and
380disabled by writing to the "cgroup.subtree_control" file::
381
382  # echo "+cpu +memory -io" > cgroup.subtree_control
383
384Only controllers which are listed in "cgroup.controllers" can be
385enabled.  When multiple operations are specified as above, either they
386all succeed or fail.  If multiple operations on the same controller
387are specified, the last one is effective.
388
389Enabling a controller in a cgroup indicates that the distribution of
390the target resource across its immediate children will be controlled.
391Consider the following sub-hierarchy.  The enabled controllers are
392listed in parentheses::
393
394  A(cpu,memory) - B(memory) - C()
395                            \ D()
396
397As A has "cpu" and "memory" enabled, A will control the distribution
398of CPU cycles and memory to its children, in this case, B.  As B has
399"memory" enabled but not "CPU", C and D will compete freely on CPU
400cycles but their division of memory available to B will be controlled.
401
402As a controller regulates the distribution of the target resource to
403the cgroup's children, enabling it creates the controller's interface
404files in the child cgroups.  In the above example, enabling "cpu" on B
405would create the "cpu." prefixed controller interface files in C and
406D.  Likewise, disabling "memory" from B would remove the "memory."
407prefixed controller interface files from C and D.  This means that the
408controller interface files - anything which doesn't start with
409"cgroup." are owned by the parent rather than the cgroup itself.
410
411
412Top-down Constraint
413~~~~~~~~~~~~~~~~~~~
414
415Resources are distributed top-down and a cgroup can further distribute
416a resource only if the resource has been distributed to it from the
417parent.  This means that all non-root "cgroup.subtree_control" files
418can only contain controllers which are enabled in the parent's
419"cgroup.subtree_control" file.  A controller can be enabled only if
420the parent has the controller enabled and a controller can't be
421disabled if one or more children have it enabled.
422
423
424No Internal Process Constraint
425~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
426
427Non-root cgroups can distribute domain resources to their children
428only when they don't have any processes of their own.  In other words,
429only domain cgroups which don't contain any processes can have domain
430controllers enabled in their "cgroup.subtree_control" files.
431
432This guarantees that, when a domain controller is looking at the part
433of the hierarchy which has it enabled, processes are always only on
434the leaves.  This rules out situations where child cgroups compete
435against internal processes of the parent.
436
437The root cgroup is exempt from this restriction.  Root contains
438processes and anonymous resource consumption which can't be associated
439with any other cgroups and requires special treatment from most
440controllers.  How resource consumption in the root cgroup is governed
441is up to each controller (for more information on this topic please
442refer to the Non-normative information section in the Controllers
443chapter).
444
445Note that the restriction doesn't get in the way if there is no
446enabled controller in the cgroup's "cgroup.subtree_control".  This is
447important as otherwise it wouldn't be possible to create children of a
448populated cgroup.  To control resource distribution of a cgroup, the
449cgroup must create children and transfer all its processes to the
450children before enabling controllers in its "cgroup.subtree_control"
451file.
452
453
454Delegation
455----------
456
457Model of Delegation
458~~~~~~~~~~~~~~~~~~~
459
460A cgroup can be delegated in two ways.  First, to a less privileged
461user by granting write access of the directory and its "cgroup.procs",
462"cgroup.threads" and "cgroup.subtree_control" files to the user.
463Second, if the "nsdelegate" mount option is set, automatically to a
464cgroup namespace on namespace creation.
465
466Because the resource control interface files in a given directory
467control the distribution of the parent's resources, the delegatee
468shouldn't be allowed to write to them.  For the first method, this is
469achieved by not granting access to these files.  For the second, the
470kernel rejects writes to all files other than "cgroup.procs" and
471"cgroup.subtree_control" on a namespace root from inside the
472namespace.
473
474The end results are equivalent for both delegation types.  Once
475delegated, the user can build sub-hierarchy under the directory,
476organize processes inside it as it sees fit and further distribute the
477resources it received from the parent.  The limits and other settings
478of all resource controllers are hierarchical and regardless of what
479happens in the delegated sub-hierarchy, nothing can escape the
480resource restrictions imposed by the parent.
481
482Currently, cgroup doesn't impose any restrictions on the number of
483cgroups in or nesting depth of a delegated sub-hierarchy; however,
484this may be limited explicitly in the future.
485
486
487Delegation Containment
488~~~~~~~~~~~~~~~~~~~~~~
489
490A delegated sub-hierarchy is contained in the sense that processes
491can't be moved into or out of the sub-hierarchy by the delegatee.
492
493For delegations to a less privileged user, this is achieved by
494requiring the following conditions for a process with a non-root euid
495to migrate a target process into a cgroup by writing its PID to the
496"cgroup.procs" file.
497
498- The writer must have write access to the "cgroup.procs" file.
499
500- The writer must have write access to the "cgroup.procs" file of the
501  common ancestor of the source and destination cgroups.
502
503The above two constraints ensure that while a delegatee may migrate
504processes around freely in the delegated sub-hierarchy it can't pull
505in from or push out to outside the sub-hierarchy.
506
507For an example, let's assume cgroups C0 and C1 have been delegated to
508user U0 who created C00, C01 under C0 and C10 under C1 as follows and
509all processes under C0 and C1 belong to U0::
510
511  ~~~~~~~~~~~~~ - C0 - C00
512  ~ cgroup    ~      \ C01
513  ~ hierarchy ~
514  ~~~~~~~~~~~~~ - C1 - C10
515
516Let's also say U0 wants to write the PID of a process which is
517currently in C10 into "C00/cgroup.procs".  U0 has write access to the
518file; however, the common ancestor of the source cgroup C10 and the
519destination cgroup C00 is above the points of delegation and U0 would
520not have write access to its "cgroup.procs" files and thus the write
521will be denied with -EACCES.
522
523For delegations to namespaces, containment is achieved by requiring
524that both the source and destination cgroups are reachable from the
525namespace of the process which is attempting the migration.  If either
526is not reachable, the migration is rejected with -ENOENT.
527
528
529Guidelines
530----------
531
532Organize Once and Control
533~~~~~~~~~~~~~~~~~~~~~~~~~
534
535Migrating a process across cgroups is a relatively expensive operation
536and stateful resources such as memory are not moved together with the
537process.  This is an explicit design decision as there often exist
538inherent trade-offs between migration and various hot paths in terms
539of synchronization cost.
540
541As such, migrating processes across cgroups frequently as a means to
542apply different resource restrictions is discouraged.  A workload
543should be assigned to a cgroup according to the system's logical and
544resource structure once on start-up.  Dynamic adjustments to resource
545distribution can be made by changing controller configuration through
546the interface files.
547
548
549Avoid Name Collisions
550~~~~~~~~~~~~~~~~~~~~~
551
552Interface files for a cgroup and its children cgroups occupy the same
553directory and it is possible to create children cgroups which collide
554with interface files.
555
556All cgroup core interface files are prefixed with "cgroup." and each
557controller's interface files are prefixed with the controller name and
558a dot.  A controller's name is composed of lower case alphabets and
559'_'s but never begins with an '_' so it can be used as the prefix
560character for collision avoidance.  Also, interface file names won't
561start or end with terms which are often used in categorizing workloads
562such as job, service, slice, unit or workload.
563
564cgroup doesn't do anything to prevent name collisions and it's the
565user's responsibility to avoid them.
566
567
568Resource Distribution Models
569============================
570
571cgroup controllers implement several resource distribution schemes
572depending on the resource type and expected use cases.  This section
573describes major schemes in use along with their expected behaviors.
574
575
576Weights
577-------
578
579A parent's resource is distributed by adding up the weights of all
580active children and giving each the fraction matching the ratio of its
581weight against the sum.  As only children which can make use of the
582resource at the moment participate in the distribution, this is
583work-conserving.  Due to the dynamic nature, this model is usually
584used for stateless resources.
585
586All weights are in the range [1, 10000] with the default at 100.  This
587allows symmetric multiplicative biases in both directions at fine
588enough granularity while staying in the intuitive range.
589
590As long as the weight is in range, all configuration combinations are
591valid and there is no reason to reject configuration changes or
592process migrations.
593
594"cpu.weight" proportionally distributes CPU cycles to active children
595and is an example of this type.
596
597
598Limits
599------
600
601A child can only consume upto the configured amount of the resource.
602Limits can be over-committed - the sum of the limits of children can
603exceed the amount of resource available to the parent.
604
605Limits are in the range [0, max] and defaults to "max", which is noop.
606
607As limits can be over-committed, all configuration combinations are
608valid and there is no reason to reject configuration changes or
609process migrations.
610
611"io.max" limits the maximum BPS and/or IOPS that a cgroup can consume
612on an IO device and is an example of this type.
613
614
615Protections
616-----------
617
618A cgroup is protected to be allocated upto the configured amount of
619the resource if the usages of all its ancestors are under their
620protected levels.  Protections can be hard guarantees or best effort
621soft boundaries.  Protections can also be over-committed in which case
622only upto the amount available to the parent is protected among
623children.
624
625Protections are in the range [0, max] and defaults to 0, which is
626noop.
627
628As protections can be over-committed, all configuration combinations
629are valid and there is no reason to reject configuration changes or
630process migrations.
631
632"memory.low" implements best-effort memory protection and is an
633example of this type.
634
635
636Allocations
637-----------
638
639A cgroup is exclusively allocated a certain amount of a finite
640resource.  Allocations can't be over-committed - the sum of the
641allocations of children can not exceed the amount of resource
642available to the parent.
643
644Allocations are in the range [0, max] and defaults to 0, which is no
645resource.
646
647As allocations can't be over-committed, some configuration
648combinations are invalid and should be rejected.  Also, if the
649resource is mandatory for execution of processes, process migrations
650may be rejected.
651
652"cpu.rt.max" hard-allocates realtime slices and is an example of this
653type.
654
655
656Interface Files
657===============
658
659Format
660------
661
662All interface files should be in one of the following formats whenever
663possible::
664
665  New-line separated values
666  (when only one value can be written at once)
667
668	VAL0\n
669	VAL1\n
670	...
671
672  Space separated values
673  (when read-only or multiple values can be written at once)
674
675	VAL0 VAL1 ...\n
676
677  Flat keyed
678
679	KEY0 VAL0\n
680	KEY1 VAL1\n
681	...
682
683  Nested keyed
684
685	KEY0 SUB_KEY0=VAL00 SUB_KEY1=VAL01...
686	KEY1 SUB_KEY0=VAL10 SUB_KEY1=VAL11...
687	...
688
689For a writable file, the format for writing should generally match
690reading; however, controllers may allow omitting later fields or
691implement restricted shortcuts for most common use cases.
692
693For both flat and nested keyed files, only the values for a single key
694can be written at a time.  For nested keyed files, the sub key pairs
695may be specified in any order and not all pairs have to be specified.
696
697
698Conventions
699-----------
700
701- Settings for a single feature should be contained in a single file.
702
703- The root cgroup should be exempt from resource control and thus
704  shouldn't have resource control interface files.  Also,
705  informational files on the root cgroup which end up showing global
706  information available elsewhere shouldn't exist.
707
708- If a controller implements weight based resource distribution, its
709  interface file should be named "weight" and have the range [1,
710  10000] with 100 as the default.  The values are chosen to allow
711  enough and symmetric bias in both directions while keeping it
712  intuitive (the default is 100%).
713
714- If a controller implements an absolute resource guarantee and/or
715  limit, the interface files should be named "min" and "max"
716  respectively.  If a controller implements best effort resource
717  guarantee and/or limit, the interface files should be named "low"
718  and "high" respectively.
719
720  In the above four control files, the special token "max" should be
721  used to represent upward infinity for both reading and writing.
722
723- If a setting has a configurable default value and keyed specific
724  overrides, the default entry should be keyed with "default" and
725  appear as the first entry in the file.
726
727  The default value can be updated by writing either "default $VAL" or
728  "$VAL".
729
730  When writing to update a specific override, "default" can be used as
731  the value to indicate removal of the override.  Override entries
732  with "default" as the value must not appear when read.
733
734  For example, a setting which is keyed by major:minor device numbers
735  with integer values may look like the following::
736
737    # cat cgroup-example-interface-file
738    default 150
739    8:0 300
740
741  The default value can be updated by::
742
743    # echo 125 > cgroup-example-interface-file
744
745  or::
746
747    # echo "default 125" > cgroup-example-interface-file
748
749  An override can be set by::
750
751    # echo "8:16 170" > cgroup-example-interface-file
752
753  and cleared by::
754
755    # echo "8:0 default" > cgroup-example-interface-file
756    # cat cgroup-example-interface-file
757    default 125
758    8:16 170
759
760- For events which are not very high frequency, an interface file
761  "events" should be created which lists event key value pairs.
762  Whenever a notifiable event happens, file modified event should be
763  generated on the file.
764
765
766Core Interface Files
767--------------------
768
769All cgroup core files are prefixed with "cgroup."
770
771  cgroup.type
772
773	A read-write single value file which exists on non-root
774	cgroups.
775
776	When read, it indicates the current type of the cgroup, which
777	can be one of the following values.
778
779	- "domain" : A normal valid domain cgroup.
780
781	- "domain threaded" : A threaded domain cgroup which is
782          serving as the root of a threaded subtree.
783
784	- "domain invalid" : A cgroup which is in an invalid state.
785	  It can't be populated or have controllers enabled.  It may
786	  be allowed to become a threaded cgroup.
787
788	- "threaded" : A threaded cgroup which is a member of a
789          threaded subtree.
790
791	A cgroup can be turned into a threaded cgroup by writing
792	"threaded" to this file.
793
794  cgroup.procs
795	A read-write new-line separated values file which exists on
796	all cgroups.
797
798	When read, it lists the PIDs of all processes which belong to
799	the cgroup one-per-line.  The PIDs are not ordered and the
800	same PID may show up more than once if the process got moved
801	to another cgroup and then back or the PID got recycled while
802	reading.
803
804	A PID can be written to migrate the process associated with
805	the PID to the cgroup.  The writer should match all of the
806	following conditions.
807
808	- It must have write access to the "cgroup.procs" file.
809
810	- It must have write access to the "cgroup.procs" file of the
811	  common ancestor of the source and destination cgroups.
812
813	When delegating a sub-hierarchy, write access to this file
814	should be granted along with the containing directory.
815
816	In a threaded cgroup, reading this file fails with EOPNOTSUPP
817	as all the processes belong to the thread root.  Writing is
818	supported and moves every thread of the process to the cgroup.
819
820  cgroup.threads
821	A read-write new-line separated values file which exists on
822	all cgroups.
823
824	When read, it lists the TIDs of all threads which belong to
825	the cgroup one-per-line.  The TIDs are not ordered and the
826	same TID may show up more than once if the thread got moved to
827	another cgroup and then back or the TID got recycled while
828	reading.
829
830	A TID can be written to migrate the thread associated with the
831	TID to the cgroup.  The writer should match all of the
832	following conditions.
833
834	- It must have write access to the "cgroup.threads" file.
835
836	- The cgroup that the thread is currently in must be in the
837          same resource domain as the destination cgroup.
838
839	- It must have write access to the "cgroup.procs" file of the
840	  common ancestor of the source and destination cgroups.
841
842	When delegating a sub-hierarchy, write access to this file
843	should be granted along with the containing directory.
844
845  cgroup.controllers
846	A read-only space separated values file which exists on all
847	cgroups.
848
849	It shows space separated list of all controllers available to
850	the cgroup.  The controllers are not ordered.
851
852  cgroup.subtree_control
853	A read-write space separated values file which exists on all
854	cgroups.  Starts out empty.
855
856	When read, it shows space separated list of the controllers
857	which are enabled to control resource distribution from the
858	cgroup to its children.
859
860	Space separated list of controllers prefixed with '+' or '-'
861	can be written to enable or disable controllers.  A controller
862	name prefixed with '+' enables the controller and '-'
863	disables.  If a controller appears more than once on the list,
864	the last one is effective.  When multiple enable and disable
865	operations are specified, either all succeed or all fail.
866
867  cgroup.events
868	A read-only flat-keyed file which exists on non-root cgroups.
869	The following entries are defined.  Unless specified
870	otherwise, a value change in this file generates a file
871	modified event.
872
873	  populated
874		1 if the cgroup or its descendants contains any live
875		processes; otherwise, 0.
876	  frozen
877		1 if the cgroup is frozen; otherwise, 0.
878
879  cgroup.max.descendants
880	A read-write single value files.  The default is "max".
881
882	Maximum allowed number of descent cgroups.
883	If the actual number of descendants is equal or larger,
884	an attempt to create a new cgroup in the hierarchy will fail.
885
886  cgroup.max.depth
887	A read-write single value files.  The default is "max".
888
889	Maximum allowed descent depth below the current cgroup.
890	If the actual descent depth is equal or larger,
891	an attempt to create a new child cgroup will fail.
892
893  cgroup.stat
894	A read-only flat-keyed file with the following entries:
895
896	  nr_descendants
897		Total number of visible descendant cgroups.
898
899	  nr_dying_descendants
900		Total number of dying descendant cgroups. A cgroup becomes
901		dying after being deleted by a user. The cgroup will remain
902		in dying state for some time undefined time (which can depend
903		on system load) before being completely destroyed.
904
905		A process can't enter a dying cgroup under any circumstances,
906		a dying cgroup can't revive.
907
908		A dying cgroup can consume system resources not exceeding
909		limits, which were active at the moment of cgroup deletion.
910
911  cgroup.freeze
912	A read-write single value file which exists on non-root cgroups.
913	Allowed values are "0" and "1". The default is "0".
914
915	Writing "1" to the file causes freezing of the cgroup and all
916	descendant cgroups. This means that all belonging processes will
917	be stopped and will not run until the cgroup will be explicitly
918	unfrozen. Freezing of the cgroup may take some time; when this action
919	is completed, the "frozen" value in the cgroup.events control file
920	will be updated to "1" and the corresponding notification will be
921	issued.
922
923	A cgroup can be frozen either by its own settings, or by settings
924	of any ancestor cgroups. If any of ancestor cgroups is frozen, the
925	cgroup will remain frozen.
926
927	Processes in the frozen cgroup can be killed by a fatal signal.
928	They also can enter and leave a frozen cgroup: either by an explicit
929	move by a user, or if freezing of the cgroup races with fork().
930	If a process is moved to a frozen cgroup, it stops. If a process is
931	moved out of a frozen cgroup, it becomes running.
932
933	Frozen status of a cgroup doesn't affect any cgroup tree operations:
934	it's possible to delete a frozen (and empty) cgroup, as well as
935	create new sub-cgroups.
936
937Controllers
938===========
939
940CPU
941---
942
943The "cpu" controllers regulates distribution of CPU cycles.  This
944controller implements weight and absolute bandwidth limit models for
945normal scheduling policy and absolute bandwidth allocation model for
946realtime scheduling policy.
947
948WARNING: cgroup2 doesn't yet support control of realtime processes and
949the cpu controller can only be enabled when all RT processes are in
950the root cgroup.  Be aware that system management software may already
951have placed RT processes into nonroot cgroups during the system boot
952process, and these processes may need to be moved to the root cgroup
953before the cpu controller can be enabled.
954
955
956CPU Interface Files
957~~~~~~~~~~~~~~~~~~~
958
959All time durations are in microseconds.
960
961  cpu.stat
962	A read-only flat-keyed file which exists on non-root cgroups.
963	This file exists whether the controller is enabled or not.
964
965	It always reports the following three stats:
966
967	- usage_usec
968	- user_usec
969	- system_usec
970
971	and the following three when the controller is enabled:
972
973	- nr_periods
974	- nr_throttled
975	- throttled_usec
976
977  cpu.weight
978	A read-write single value file which exists on non-root
979	cgroups.  The default is "100".
980
981	The weight in the range [1, 10000].
982
983  cpu.weight.nice
984	A read-write single value file which exists on non-root
985	cgroups.  The default is "0".
986
987	The nice value is in the range [-20, 19].
988
989	This interface file is an alternative interface for
990	"cpu.weight" and allows reading and setting weight using the
991	same values used by nice(2).  Because the range is smaller and
992	granularity is coarser for the nice values, the read value is
993	the closest approximation of the current weight.
994
995  cpu.max
996	A read-write two value file which exists on non-root cgroups.
997	The default is "max 100000".
998
999	The maximum bandwidth limit.  It's in the following format::
1000
1001	  $MAX $PERIOD
1002
1003	which indicates that the group may consume upto $MAX in each
1004	$PERIOD duration.  "max" for $MAX indicates no limit.  If only
1005	one number is written, $MAX is updated.
1006
1007  cpu.pressure
1008	A read-only nested-key file which exists on non-root cgroups.
1009
1010	Shows pressure stall information for CPU. See
1011	Documentation/accounting/psi.txt for details.
1012
1013
1014Memory
1015------
1016
1017The "memory" controller regulates distribution of memory.  Memory is
1018stateful and implements both limit and protection models.  Due to the
1019intertwining between memory usage and reclaim pressure and the
1020stateful nature of memory, the distribution model is relatively
1021complex.
1022
1023While not completely water-tight, all major memory usages by a given
1024cgroup are tracked so that the total memory consumption can be
1025accounted and controlled to a reasonable extent.  Currently, the
1026following types of memory usages are tracked.
1027
1028- Userland memory - page cache and anonymous memory.
1029
1030- Kernel data structures such as dentries and inodes.
1031
1032- TCP socket buffers.
1033
1034The above list may expand in the future for better coverage.
1035
1036
1037Memory Interface Files
1038~~~~~~~~~~~~~~~~~~~~~~
1039
1040All memory amounts are in bytes.  If a value which is not aligned to
1041PAGE_SIZE is written, the value may be rounded up to the closest
1042PAGE_SIZE multiple when read back.
1043
1044  memory.current
1045	A read-only single value file which exists on non-root
1046	cgroups.
1047
1048	The total amount of memory currently being used by the cgroup
1049	and its descendants.
1050
1051  memory.min
1052	A read-write single value file which exists on non-root
1053	cgroups.  The default is "0".
1054
1055	Hard memory protection.  If the memory usage of a cgroup
1056	is within its effective min boundary, the cgroup's memory
1057	won't be reclaimed under any conditions. If there is no
1058	unprotected reclaimable memory available, OOM killer
1059	is invoked.
1060
1061       Effective min boundary is limited by memory.min values of
1062	all ancestor cgroups. If there is memory.min overcommitment
1063	(child cgroup or cgroups are requiring more protected memory
1064	than parent will allow), then each child cgroup will get
1065	the part of parent's protection proportional to its
1066	actual memory usage below memory.min.
1067
1068	Putting more memory than generally available under this
1069	protection is discouraged and may lead to constant OOMs.
1070
1071	If a memory cgroup is not populated with processes,
1072	its memory.min is ignored.
1073
1074  memory.low
1075	A read-write single value file which exists on non-root
1076	cgroups.  The default is "0".
1077
1078	Best-effort memory protection.  If the memory usage of a
1079	cgroup is within its effective low boundary, the cgroup's
1080	memory won't be reclaimed unless memory can be reclaimed
1081	from unprotected cgroups.
1082
1083	Effective low boundary is limited by memory.low values of
1084	all ancestor cgroups. If there is memory.low overcommitment
1085	(child cgroup or cgroups are requiring more protected memory
1086	than parent will allow), then each child cgroup will get
1087	the part of parent's protection proportional to its
1088	actual memory usage below memory.low.
1089
1090	Putting more memory than generally available under this
1091	protection is discouraged.
1092
1093  memory.high
1094	A read-write single value file which exists on non-root
1095	cgroups.  The default is "max".
1096
1097	Memory usage throttle limit.  This is the main mechanism to
1098	control memory usage of a cgroup.  If a cgroup's usage goes
1099	over the high boundary, the processes of the cgroup are
1100	throttled and put under heavy reclaim pressure.
1101
1102	Going over the high limit never invokes the OOM killer and
1103	under extreme conditions the limit may be breached.
1104
1105  memory.max
1106	A read-write single value file which exists on non-root
1107	cgroups.  The default is "max".
1108
1109	Memory usage hard limit.  This is the final protection
1110	mechanism.  If a cgroup's memory usage reaches this limit and
1111	can't be reduced, the OOM killer is invoked in the cgroup.
1112	Under certain circumstances, the usage may go over the limit
1113	temporarily.
1114
1115	This is the ultimate protection mechanism.  As long as the
1116	high limit is used and monitored properly, this limit's
1117	utility is limited to providing the final safety net.
1118
1119  memory.oom.group
1120	A read-write single value file which exists on non-root
1121	cgroups.  The default value is "0".
1122
1123	Determines whether the cgroup should be treated as
1124	an indivisible workload by the OOM killer. If set,
1125	all tasks belonging to the cgroup or to its descendants
1126	(if the memory cgroup is not a leaf cgroup) are killed
1127	together or not at all. This can be used to avoid
1128	partial kills to guarantee workload integrity.
1129
1130	Tasks with the OOM protection (oom_score_adj set to -1000)
1131	are treated as an exception and are never killed.
1132
1133	If the OOM killer is invoked in a cgroup, it's not going
1134	to kill any tasks outside of this cgroup, regardless
1135	memory.oom.group values of ancestor cgroups.
1136
1137  memory.events
1138	A read-only flat-keyed file which exists on non-root cgroups.
1139	The following entries are defined.  Unless specified
1140	otherwise, a value change in this file generates a file
1141	modified event.
1142
1143	  low
1144		The number of times the cgroup is reclaimed due to
1145		high memory pressure even though its usage is under
1146		the low boundary.  This usually indicates that the low
1147		boundary is over-committed.
1148
1149	  high
1150		The number of times processes of the cgroup are
1151		throttled and routed to perform direct memory reclaim
1152		because the high memory boundary was exceeded.  For a
1153		cgroup whose memory usage is capped by the high limit
1154		rather than global memory pressure, this event's
1155		occurrences are expected.
1156
1157	  max
1158		The number of times the cgroup's memory usage was
1159		about to go over the max boundary.  If direct reclaim
1160		fails to bring it down, the cgroup goes to OOM state.
1161
1162	  oom
1163		The number of time the cgroup's memory usage was
1164		reached the limit and allocation was about to fail.
1165
1166		Depending on context result could be invocation of OOM
1167		killer and retrying allocation or failing allocation.
1168
1169		Failed allocation in its turn could be returned into
1170		userspace as -ENOMEM or silently ignored in cases like
1171		disk readahead.  For now OOM in memory cgroup kills
1172		tasks iff shortage has happened inside page fault.
1173
1174		This event is not raised if the OOM killer is not
1175		considered as an option, e.g. for failed high-order
1176		allocations.
1177
1178	  oom_kill
1179		The number of processes belonging to this cgroup
1180		killed by any kind of OOM killer.
1181
1182  memory.stat
1183	A read-only flat-keyed file which exists on non-root cgroups.
1184
1185	This breaks down the cgroup's memory footprint into different
1186	types of memory, type-specific details, and other information
1187	on the state and past events of the memory management system.
1188
1189	All memory amounts are in bytes.
1190
1191	The entries are ordered to be human readable, and new entries
1192	can show up in the middle. Don't rely on items remaining in a
1193	fixed position; use the keys to look up specific values!
1194
1195	  anon
1196		Amount of memory used in anonymous mappings such as
1197		brk(), sbrk(), and mmap(MAP_ANONYMOUS)
1198
1199	  file
1200		Amount of memory used to cache filesystem data,
1201		including tmpfs and shared memory.
1202
1203	  kernel_stack
1204		Amount of memory allocated to kernel stacks.
1205
1206	  slab
1207		Amount of memory used for storing in-kernel data
1208		structures.
1209
1210	  sock
1211		Amount of memory used in network transmission buffers
1212
1213	  shmem
1214		Amount of cached filesystem data that is swap-backed,
1215		such as tmpfs, shm segments, shared anonymous mmap()s
1216
1217	  file_mapped
1218		Amount of cached filesystem data mapped with mmap()
1219
1220	  file_dirty
1221		Amount of cached filesystem data that was modified but
1222		not yet written back to disk
1223
1224	  file_writeback
1225		Amount of cached filesystem data that was modified and
1226		is currently being written back to disk
1227
1228	  anon_thp
1229		Amount of memory used in anonymous mappings backed by
1230		transparent hugepages
1231
1232	  inactive_anon, active_anon, inactive_file, active_file, unevictable
1233		Amount of memory, swap-backed and filesystem-backed,
1234		on the internal memory management lists used by the
1235		page reclaim algorithm
1236
1237	  slab_reclaimable
1238		Part of "slab" that might be reclaimed, such as
1239		dentries and inodes.
1240
1241	  slab_unreclaimable
1242		Part of "slab" that cannot be reclaimed on memory
1243		pressure.
1244
1245	  pgfault
1246		Total number of page faults incurred
1247
1248	  pgmajfault
1249		Number of major page faults incurred
1250
1251	  workingset_refault
1252
1253		Number of refaults of previously evicted pages
1254
1255	  workingset_activate
1256
1257		Number of refaulted pages that were immediately activated
1258
1259	  workingset_nodereclaim
1260
1261		Number of times a shadow node has been reclaimed
1262
1263	  pgrefill
1264
1265		Amount of scanned pages (in an active LRU list)
1266
1267	  pgscan
1268
1269		Amount of scanned pages (in an inactive LRU list)
1270
1271	  pgsteal
1272
1273		Amount of reclaimed pages
1274
1275	  pgactivate
1276
1277		Amount of pages moved to the active LRU list
1278
1279	  pgdeactivate
1280
1281		Amount of pages moved to the inactive LRU lis
1282
1283	  pglazyfree
1284
1285		Amount of pages postponed to be freed under memory pressure
1286
1287	  pglazyfreed
1288
1289		Amount of reclaimed lazyfree pages
1290
1291	  thp_fault_alloc
1292
1293		Number of transparent hugepages which were allocated to satisfy
1294		a page fault, including COW faults. This counter is not present
1295		when CONFIG_TRANSPARENT_HUGEPAGE is not set.
1296
1297	  thp_collapse_alloc
1298
1299		Number of transparent hugepages which were allocated to allow
1300		collapsing an existing range of pages. This counter is not
1301		present when CONFIG_TRANSPARENT_HUGEPAGE is not set.
1302
1303  memory.swap.current
1304	A read-only single value file which exists on non-root
1305	cgroups.
1306
1307	The total amount of swap currently being used by the cgroup
1308	and its descendants.
1309
1310  memory.swap.max
1311	A read-write single value file which exists on non-root
1312	cgroups.  The default is "max".
1313
1314	Swap usage hard limit.  If a cgroup's swap usage reaches this
1315	limit, anonymous memory of the cgroup will not be swapped out.
1316
1317  memory.swap.events
1318	A read-only flat-keyed file which exists on non-root cgroups.
1319	The following entries are defined.  Unless specified
1320	otherwise, a value change in this file generates a file
1321	modified event.
1322
1323	  max
1324		The number of times the cgroup's swap usage was about
1325		to go over the max boundary and swap allocation
1326		failed.
1327
1328	  fail
1329		The number of times swap allocation failed either
1330		because of running out of swap system-wide or max
1331		limit.
1332
1333	When reduced under the current usage, the existing swap
1334	entries are reclaimed gradually and the swap usage may stay
1335	higher than the limit for an extended period of time.  This
1336	reduces the impact on the workload and memory management.
1337
1338  memory.pressure
1339	A read-only nested-key file which exists on non-root cgroups.
1340
1341	Shows pressure stall information for memory. See
1342	Documentation/accounting/psi.txt for details.
1343
1344
1345Usage Guidelines
1346~~~~~~~~~~~~~~~~
1347
1348"memory.high" is the main mechanism to control memory usage.
1349Over-committing on high limit (sum of high limits > available memory)
1350and letting global memory pressure to distribute memory according to
1351usage is a viable strategy.
1352
1353Because breach of the high limit doesn't trigger the OOM killer but
1354throttles the offending cgroup, a management agent has ample
1355opportunities to monitor and take appropriate actions such as granting
1356more memory or terminating the workload.
1357
1358Determining whether a cgroup has enough memory is not trivial as
1359memory usage doesn't indicate whether the workload can benefit from
1360more memory.  For example, a workload which writes data received from
1361network to a file can use all available memory but can also operate as
1362performant with a small amount of memory.  A measure of memory
1363pressure - how much the workload is being impacted due to lack of
1364memory - is necessary to determine whether a workload needs more
1365memory; unfortunately, memory pressure monitoring mechanism isn't
1366implemented yet.
1367
1368
1369Memory Ownership
1370~~~~~~~~~~~~~~~~
1371
1372A memory area is charged to the cgroup which instantiated it and stays
1373charged to the cgroup until the area is released.  Migrating a process
1374to a different cgroup doesn't move the memory usages that it
1375instantiated while in the previous cgroup to the new cgroup.
1376
1377A memory area may be used by processes belonging to different cgroups.
1378To which cgroup the area will be charged is in-deterministic; however,
1379over time, the memory area is likely to end up in a cgroup which has
1380enough memory allowance to avoid high reclaim pressure.
1381
1382If a cgroup sweeps a considerable amount of memory which is expected
1383to be accessed repeatedly by other cgroups, it may make sense to use
1384POSIX_FADV_DONTNEED to relinquish the ownership of memory areas
1385belonging to the affected files to ensure correct memory ownership.
1386
1387
1388IO
1389--
1390
1391The "io" controller regulates the distribution of IO resources.  This
1392controller implements both weight based and absolute bandwidth or IOPS
1393limit distribution; however, weight based distribution is available
1394only if cfq-iosched is in use and neither scheme is available for
1395blk-mq devices.
1396
1397
1398IO Interface Files
1399~~~~~~~~~~~~~~~~~~
1400
1401  io.stat
1402	A read-only nested-keyed file which exists on non-root
1403	cgroups.
1404
1405	Lines are keyed by $MAJ:$MIN device numbers and not ordered.
1406	The following nested keys are defined.
1407
1408	  ======	=====================
1409	  rbytes	Bytes read
1410	  wbytes	Bytes written
1411	  rios		Number of read IOs
1412	  wios		Number of write IOs
1413	  dbytes	Bytes discarded
1414	  dios		Number of discard IOs
1415	  ======	=====================
1416
1417	An example read output follows:
1418
1419	  8:16 rbytes=1459200 wbytes=314773504 rios=192 wios=353 dbytes=0 dios=0
1420	  8:0 rbytes=90430464 wbytes=299008000 rios=8950 wios=1252 dbytes=50331648 dios=3021
1421
1422  io.weight
1423	A read-write flat-keyed file which exists on non-root cgroups.
1424	The default is "default 100".
1425
1426	The first line is the default weight applied to devices
1427	without specific override.  The rest are overrides keyed by
1428	$MAJ:$MIN device numbers and not ordered.  The weights are in
1429	the range [1, 10000] and specifies the relative amount IO time
1430	the cgroup can use in relation to its siblings.
1431
1432	The default weight can be updated by writing either "default
1433	$WEIGHT" or simply "$WEIGHT".  Overrides can be set by writing
1434	"$MAJ:$MIN $WEIGHT" and unset by writing "$MAJ:$MIN default".
1435
1436	An example read output follows::
1437
1438	  default 100
1439	  8:16 200
1440	  8:0 50
1441
1442  io.max
1443	A read-write nested-keyed file which exists on non-root
1444	cgroups.
1445
1446	BPS and IOPS based IO limit.  Lines are keyed by $MAJ:$MIN
1447	device numbers and not ordered.  The following nested keys are
1448	defined.
1449
1450	  =====		==================================
1451	  rbps		Max read bytes per second
1452	  wbps		Max write bytes per second
1453	  riops		Max read IO operations per second
1454	  wiops		Max write IO operations per second
1455	  =====		==================================
1456
1457	When writing, any number of nested key-value pairs can be
1458	specified in any order.  "max" can be specified as the value
1459	to remove a specific limit.  If the same key is specified
1460	multiple times, the outcome is undefined.
1461
1462	BPS and IOPS are measured in each IO direction and IOs are
1463	delayed if limit is reached.  Temporary bursts are allowed.
1464
1465	Setting read limit at 2M BPS and write at 120 IOPS for 8:16::
1466
1467	  echo "8:16 rbps=2097152 wiops=120" > io.max
1468
1469	Reading returns the following::
1470
1471	  8:16 rbps=2097152 wbps=max riops=max wiops=120
1472
1473	Write IOPS limit can be removed by writing the following::
1474
1475	  echo "8:16 wiops=max" > io.max
1476
1477	Reading now returns the following::
1478
1479	  8:16 rbps=2097152 wbps=max riops=max wiops=max
1480
1481  io.pressure
1482	A read-only nested-key file which exists on non-root cgroups.
1483
1484	Shows pressure stall information for IO. See
1485	Documentation/accounting/psi.txt for details.
1486
1487
1488Writeback
1489~~~~~~~~~
1490
1491Page cache is dirtied through buffered writes and shared mmaps and
1492written asynchronously to the backing filesystem by the writeback
1493mechanism.  Writeback sits between the memory and IO domains and
1494regulates the proportion of dirty memory by balancing dirtying and
1495write IOs.
1496
1497The io controller, in conjunction with the memory controller,
1498implements control of page cache writeback IOs.  The memory controller
1499defines the memory domain that dirty memory ratio is calculated and
1500maintained for and the io controller defines the io domain which
1501writes out dirty pages for the memory domain.  Both system-wide and
1502per-cgroup dirty memory states are examined and the more restrictive
1503of the two is enforced.
1504
1505cgroup writeback requires explicit support from the underlying
1506filesystem.  Currently, cgroup writeback is implemented on ext2, ext4
1507and btrfs.  On other filesystems, all writeback IOs are attributed to
1508the root cgroup.
1509
1510There are inherent differences in memory and writeback management
1511which affects how cgroup ownership is tracked.  Memory is tracked per
1512page while writeback per inode.  For the purpose of writeback, an
1513inode is assigned to a cgroup and all IO requests to write dirty pages
1514from the inode are attributed to that cgroup.
1515
1516As cgroup ownership for memory is tracked per page, there can be pages
1517which are associated with different cgroups than the one the inode is
1518associated with.  These are called foreign pages.  The writeback
1519constantly keeps track of foreign pages and, if a particular foreign
1520cgroup becomes the majority over a certain period of time, switches
1521the ownership of the inode to that cgroup.
1522
1523While this model is enough for most use cases where a given inode is
1524mostly dirtied by a single cgroup even when the main writing cgroup
1525changes over time, use cases where multiple cgroups write to a single
1526inode simultaneously are not supported well.  In such circumstances, a
1527significant portion of IOs are likely to be attributed incorrectly.
1528As memory controller assigns page ownership on the first use and
1529doesn't update it until the page is released, even if writeback
1530strictly follows page ownership, multiple cgroups dirtying overlapping
1531areas wouldn't work as expected.  It's recommended to avoid such usage
1532patterns.
1533
1534The sysctl knobs which affect writeback behavior are applied to cgroup
1535writeback as follows.
1536
1537  vm.dirty_background_ratio, vm.dirty_ratio
1538	These ratios apply the same to cgroup writeback with the
1539	amount of available memory capped by limits imposed by the
1540	memory controller and system-wide clean memory.
1541
1542  vm.dirty_background_bytes, vm.dirty_bytes
1543	For cgroup writeback, this is calculated into ratio against
1544	total available memory and applied the same way as
1545	vm.dirty[_background]_ratio.
1546
1547
1548IO Latency
1549~~~~~~~~~~
1550
1551This is a cgroup v2 controller for IO workload protection.  You provide a group
1552with a latency target, and if the average latency exceeds that target the
1553controller will throttle any peers that have a lower latency target than the
1554protected workload.
1555
1556The limits are only applied at the peer level in the hierarchy.  This means that
1557in the diagram below, only groups A, B, and C will influence each other, and
1558groups D and F will influence each other.  Group G will influence nobody::
1559
1560			[root]
1561		/	   |		\
1562		A	   B		C
1563	       /  \        |
1564	      D    F	   G
1565
1566
1567So the ideal way to configure this is to set io.latency in groups A, B, and C.
1568Generally you do not want to set a value lower than the latency your device
1569supports.  Experiment to find the value that works best for your workload.
1570Start at higher than the expected latency for your device and watch the
1571avg_lat value in io.stat for your workload group to get an idea of the
1572latency you see during normal operation.  Use the avg_lat value as a basis for
1573your real setting, setting at 10-15% higher than the value in io.stat.
1574
1575How IO Latency Throttling Works
1576~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1577
1578io.latency is work conserving; so as long as everybody is meeting their latency
1579target the controller doesn't do anything.  Once a group starts missing its
1580target it begins throttling any peer group that has a higher target than itself.
1581This throttling takes 2 forms:
1582
1583- Queue depth throttling.  This is the number of outstanding IO's a group is
1584  allowed to have.  We will clamp down relatively quickly, starting at no limit
1585  and going all the way down to 1 IO at a time.
1586
1587- Artificial delay induction.  There are certain types of IO that cannot be
1588  throttled without possibly adversely affecting higher priority groups.  This
1589  includes swapping and metadata IO.  These types of IO are allowed to occur
1590  normally, however they are "charged" to the originating group.  If the
1591  originating group is being throttled you will see the use_delay and delay
1592  fields in io.stat increase.  The delay value is how many microseconds that are
1593  being added to any process that runs in this group.  Because this number can
1594  grow quite large if there is a lot of swapping or metadata IO occurring we
1595  limit the individual delay events to 1 second at a time.
1596
1597Once the victimized group starts meeting its latency target again it will start
1598unthrottling any peer groups that were throttled previously.  If the victimized
1599group simply stops doing IO the global counter will unthrottle appropriately.
1600
1601IO Latency Interface Files
1602~~~~~~~~~~~~~~~~~~~~~~~~~~
1603
1604  io.latency
1605	This takes a similar format as the other controllers.
1606
1607		"MAJOR:MINOR target=<target time in microseconds"
1608
1609  io.stat
1610	If the controller is enabled you will see extra stats in io.stat in
1611	addition to the normal ones.
1612
1613	  depth
1614		This is the current queue depth for the group.
1615
1616	  avg_lat
1617		This is an exponential moving average with a decay rate of 1/exp
1618		bound by the sampling interval.  The decay rate interval can be
1619		calculated by multiplying the win value in io.stat by the
1620		corresponding number of samples based on the win value.
1621
1622	  win
1623		The sampling window size in milliseconds.  This is the minimum
1624		duration of time between evaluation events.  Windows only elapse
1625		with IO activity.  Idle periods extend the most recent window.
1626
1627PID
1628---
1629
1630The process number controller is used to allow a cgroup to stop any
1631new tasks from being fork()'d or clone()'d after a specified limit is
1632reached.
1633
1634The number of tasks in a cgroup can be exhausted in ways which other
1635controllers cannot prevent, thus warranting its own controller.  For
1636example, a fork bomb is likely to exhaust the number of tasks before
1637hitting memory restrictions.
1638
1639Note that PIDs used in this controller refer to TIDs, process IDs as
1640used by the kernel.
1641
1642
1643PID Interface Files
1644~~~~~~~~~~~~~~~~~~~
1645
1646  pids.max
1647	A read-write single value file which exists on non-root
1648	cgroups.  The default is "max".
1649
1650	Hard limit of number of processes.
1651
1652  pids.current
1653	A read-only single value file which exists on all cgroups.
1654
1655	The number of processes currently in the cgroup and its
1656	descendants.
1657
1658Organisational operations are not blocked by cgroup policies, so it is
1659possible to have pids.current > pids.max.  This can be done by either
1660setting the limit to be smaller than pids.current, or attaching enough
1661processes to the cgroup such that pids.current is larger than
1662pids.max.  However, it is not possible to violate a cgroup PID policy
1663through fork() or clone(). These will return -EAGAIN if the creation
1664of a new process would cause a cgroup policy to be violated.
1665
1666
1667Cpuset
1668------
1669
1670The "cpuset" controller provides a mechanism for constraining
1671the CPU and memory node placement of tasks to only the resources
1672specified in the cpuset interface files in a task's current cgroup.
1673This is especially valuable on large NUMA systems where placing jobs
1674on properly sized subsets of the systems with careful processor and
1675memory placement to reduce cross-node memory access and contention
1676can improve overall system performance.
1677
1678The "cpuset" controller is hierarchical.  That means the controller
1679cannot use CPUs or memory nodes not allowed in its parent.
1680
1681
1682Cpuset Interface Files
1683~~~~~~~~~~~~~~~~~~~~~~
1684
1685  cpuset.cpus
1686	A read-write multiple values file which exists on non-root
1687	cpuset-enabled cgroups.
1688
1689	It lists the requested CPUs to be used by tasks within this
1690	cgroup.  The actual list of CPUs to be granted, however, is
1691	subjected to constraints imposed by its parent and can differ
1692	from the requested CPUs.
1693
1694	The CPU numbers are comma-separated numbers or ranges.
1695	For example:
1696
1697	  # cat cpuset.cpus
1698	  0-4,6,8-10
1699
1700	An empty value indicates that the cgroup is using the same
1701	setting as the nearest cgroup ancestor with a non-empty
1702	"cpuset.cpus" or all the available CPUs if none is found.
1703
1704	The value of "cpuset.cpus" stays constant until the next update
1705	and won't be affected by any CPU hotplug events.
1706
1707  cpuset.cpus.effective
1708	A read-only multiple values file which exists on all
1709	cpuset-enabled cgroups.
1710
1711	It lists the onlined CPUs that are actually granted to this
1712	cgroup by its parent.  These CPUs are allowed to be used by
1713	tasks within the current cgroup.
1714
1715	If "cpuset.cpus" is empty, the "cpuset.cpus.effective" file shows
1716	all the CPUs from the parent cgroup that can be available to
1717	be used by this cgroup.  Otherwise, it should be a subset of
1718	"cpuset.cpus" unless none of the CPUs listed in "cpuset.cpus"
1719	can be granted.  In this case, it will be treated just like an
1720	empty "cpuset.cpus".
1721
1722	Its value will be affected by CPU hotplug events.
1723
1724  cpuset.mems
1725	A read-write multiple values file which exists on non-root
1726	cpuset-enabled cgroups.
1727
1728	It lists the requested memory nodes to be used by tasks within
1729	this cgroup.  The actual list of memory nodes granted, however,
1730	is subjected to constraints imposed by its parent and can differ
1731	from the requested memory nodes.
1732
1733	The memory node numbers are comma-separated numbers or ranges.
1734	For example:
1735
1736	  # cat cpuset.mems
1737	  0-1,3
1738
1739	An empty value indicates that the cgroup is using the same
1740	setting as the nearest cgroup ancestor with a non-empty
1741	"cpuset.mems" or all the available memory nodes if none
1742	is found.
1743
1744	The value of "cpuset.mems" stays constant until the next update
1745	and won't be affected by any memory nodes hotplug events.
1746
1747  cpuset.mems.effective
1748	A read-only multiple values file which exists on all
1749	cpuset-enabled cgroups.
1750
1751	It lists the onlined memory nodes that are actually granted to
1752	this cgroup by its parent. These memory nodes are allowed to
1753	be used by tasks within the current cgroup.
1754
1755	If "cpuset.mems" is empty, it shows all the memory nodes from the
1756	parent cgroup that will be available to be used by this cgroup.
1757	Otherwise, it should be a subset of "cpuset.mems" unless none of
1758	the memory nodes listed in "cpuset.mems" can be granted.  In this
1759	case, it will be treated just like an empty "cpuset.mems".
1760
1761	Its value will be affected by memory nodes hotplug events.
1762
1763  cpuset.cpus.partition
1764	A read-write single value file which exists on non-root
1765	cpuset-enabled cgroups.  This flag is owned by the parent cgroup
1766	and is not delegatable.
1767
1768        It accepts only the following input values when written to.
1769
1770        "root"   - a paritition root
1771        "member" - a non-root member of a partition
1772
1773	When set to be a partition root, the current cgroup is the
1774	root of a new partition or scheduling domain that comprises
1775	itself and all its descendants except those that are separate
1776	partition roots themselves and their descendants.  The root
1777	cgroup is always a partition root.
1778
1779	There are constraints on where a partition root can be set.
1780	It can only be set in a cgroup if all the following conditions
1781	are true.
1782
1783	1) The "cpuset.cpus" is not empty and the list of CPUs are
1784	   exclusive, i.e. they are not shared by any of its siblings.
1785	2) The parent cgroup is a partition root.
1786	3) The "cpuset.cpus" is also a proper subset of the parent's
1787	   "cpuset.cpus.effective".
1788	4) There is no child cgroups with cpuset enabled.  This is for
1789	   eliminating corner cases that have to be handled if such a
1790	   condition is allowed.
1791
1792	Setting it to partition root will take the CPUs away from the
1793	effective CPUs of the parent cgroup.  Once it is set, this
1794	file cannot be reverted back to "member" if there are any child
1795	cgroups with cpuset enabled.
1796
1797	A parent partition cannot distribute all its CPUs to its
1798	child partitions.  There must be at least one cpu left in the
1799	parent partition.
1800
1801	Once becoming a partition root, changes to "cpuset.cpus" is
1802	generally allowed as long as the first condition above is true,
1803	the change will not take away all the CPUs from the parent
1804	partition and the new "cpuset.cpus" value is a superset of its
1805	children's "cpuset.cpus" values.
1806
1807	Sometimes, external factors like changes to ancestors'
1808	"cpuset.cpus" or cpu hotplug can cause the state of the partition
1809	root to change.  On read, the "cpuset.sched.partition" file
1810	can show the following values.
1811
1812	"member"       Non-root member of a partition
1813	"root"         Partition root
1814	"root invalid" Invalid partition root
1815
1816	It is a partition root if the first 2 partition root conditions
1817	above are true and at least one CPU from "cpuset.cpus" is
1818	granted by the parent cgroup.
1819
1820	A partition root can become invalid if none of CPUs requested
1821	in "cpuset.cpus" can be granted by the parent cgroup or the
1822	parent cgroup is no longer a partition root itself.  In this
1823	case, it is not a real partition even though the restriction
1824	of the first partition root condition above will still apply.
1825	The cpu affinity of all the tasks in the cgroup will then be
1826	associated with CPUs in the nearest ancestor partition.
1827
1828	An invalid partition root can be transitioned back to a
1829	real partition root if at least one of the requested CPUs
1830	can now be granted by its parent.  In this case, the cpu
1831	affinity of all the tasks in the formerly invalid partition
1832	will be associated to the CPUs of the newly formed partition.
1833	Changing the partition state of an invalid partition root to
1834	"member" is always allowed even if child cpusets are present.
1835
1836
1837Device controller
1838-----------------
1839
1840Device controller manages access to device files. It includes both
1841creation of new device files (using mknod), and access to the
1842existing device files.
1843
1844Cgroup v2 device controller has no interface files and is implemented
1845on top of cgroup BPF. To control access to device files, a user may
1846create bpf programs of the BPF_CGROUP_DEVICE type and attach them
1847to cgroups. On an attempt to access a device file, corresponding
1848BPF programs will be executed, and depending on the return value
1849the attempt will succeed or fail with -EPERM.
1850
1851A BPF_CGROUP_DEVICE program takes a pointer to the bpf_cgroup_dev_ctx
1852structure, which describes the device access attempt: access type
1853(mknod/read/write) and device (type, major and minor numbers).
1854If the program returns 0, the attempt fails with -EPERM, otherwise
1855it succeeds.
1856
1857An example of BPF_CGROUP_DEVICE program may be found in the kernel
1858source tree in the tools/testing/selftests/bpf/dev_cgroup.c file.
1859
1860
1861RDMA
1862----
1863
1864The "rdma" controller regulates the distribution and accounting of
1865of RDMA resources.
1866
1867RDMA Interface Files
1868~~~~~~~~~~~~~~~~~~~~
1869
1870  rdma.max
1871	A readwrite nested-keyed file that exists for all the cgroups
1872	except root that describes current configured resource limit
1873	for a RDMA/IB device.
1874
1875	Lines are keyed by device name and are not ordered.
1876	Each line contains space separated resource name and its configured
1877	limit that can be distributed.
1878
1879	The following nested keys are defined.
1880
1881	  ==========	=============================
1882	  hca_handle	Maximum number of HCA Handles
1883	  hca_object 	Maximum number of HCA Objects
1884	  ==========	=============================
1885
1886	An example for mlx4 and ocrdma device follows::
1887
1888	  mlx4_0 hca_handle=2 hca_object=2000
1889	  ocrdma1 hca_handle=3 hca_object=max
1890
1891  rdma.current
1892	A read-only file that describes current resource usage.
1893	It exists for all the cgroup except root.
1894
1895	An example for mlx4 and ocrdma device follows::
1896
1897	  mlx4_0 hca_handle=1 hca_object=20
1898	  ocrdma1 hca_handle=1 hca_object=23
1899
1900
1901Misc
1902----
1903
1904perf_event
1905~~~~~~~~~~
1906
1907perf_event controller, if not mounted on a legacy hierarchy, is
1908automatically enabled on the v2 hierarchy so that perf events can
1909always be filtered by cgroup v2 path.  The controller can still be
1910moved to a legacy hierarchy after v2 hierarchy is populated.
1911
1912
1913Non-normative information
1914-------------------------
1915
1916This section contains information that isn't considered to be a part of
1917the stable kernel API and so is subject to change.
1918
1919
1920CPU controller root cgroup process behaviour
1921~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1922
1923When distributing CPU cycles in the root cgroup each thread in this
1924cgroup is treated as if it was hosted in a separate child cgroup of the
1925root cgroup. This child cgroup weight is dependent on its thread nice
1926level.
1927
1928For details of this mapping see sched_prio_to_weight array in
1929kernel/sched/core.c file (values from this array should be scaled
1930appropriately so the neutral - nice 0 - value is 100 instead of 1024).
1931
1932
1933IO controller root cgroup process behaviour
1934~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1935
1936Root cgroup processes are hosted in an implicit leaf child node.
1937When distributing IO resources this implicit child node is taken into
1938account as if it was a normal child cgroup of the root cgroup with a
1939weight value of 200.
1940
1941
1942Namespace
1943=========
1944
1945Basics
1946------
1947
1948cgroup namespace provides a mechanism to virtualize the view of the
1949"/proc/$PID/cgroup" file and cgroup mounts.  The CLONE_NEWCGROUP clone
1950flag can be used with clone(2) and unshare(2) to create a new cgroup
1951namespace.  The process running inside the cgroup namespace will have
1952its "/proc/$PID/cgroup" output restricted to cgroupns root.  The
1953cgroupns root is the cgroup of the process at the time of creation of
1954the cgroup namespace.
1955
1956Without cgroup namespace, the "/proc/$PID/cgroup" file shows the
1957complete path of the cgroup of a process.  In a container setup where
1958a set of cgroups and namespaces are intended to isolate processes the
1959"/proc/$PID/cgroup" file may leak potential system level information
1960to the isolated processes.  For Example::
1961
1962  # cat /proc/self/cgroup
1963  0::/batchjobs/container_id1
1964
1965The path '/batchjobs/container_id1' can be considered as system-data
1966and undesirable to expose to the isolated processes.  cgroup namespace
1967can be used to restrict visibility of this path.  For example, before
1968creating a cgroup namespace, one would see::
1969
1970  # ls -l /proc/self/ns/cgroup
1971  lrwxrwxrwx 1 root root 0 2014-07-15 10:37 /proc/self/ns/cgroup -> cgroup:[4026531835]
1972  # cat /proc/self/cgroup
1973  0::/batchjobs/container_id1
1974
1975After unsharing a new namespace, the view changes::
1976
1977  # ls -l /proc/self/ns/cgroup
1978  lrwxrwxrwx 1 root root 0 2014-07-15 10:35 /proc/self/ns/cgroup -> cgroup:[4026532183]
1979  # cat /proc/self/cgroup
1980  0::/
1981
1982When some thread from a multi-threaded process unshares its cgroup
1983namespace, the new cgroupns gets applied to the entire process (all
1984the threads).  This is natural for the v2 hierarchy; however, for the
1985legacy hierarchies, this may be unexpected.
1986
1987A cgroup namespace is alive as long as there are processes inside or
1988mounts pinning it.  When the last usage goes away, the cgroup
1989namespace is destroyed.  The cgroupns root and the actual cgroups
1990remain.
1991
1992
1993The Root and Views
1994------------------
1995
1996The 'cgroupns root' for a cgroup namespace is the cgroup in which the
1997process calling unshare(2) is running.  For example, if a process in
1998/batchjobs/container_id1 cgroup calls unshare, cgroup
1999/batchjobs/container_id1 becomes the cgroupns root.  For the
2000init_cgroup_ns, this is the real root ('/') cgroup.
2001
2002The cgroupns root cgroup does not change even if the namespace creator
2003process later moves to a different cgroup::
2004
2005  # ~/unshare -c # unshare cgroupns in some cgroup
2006  # cat /proc/self/cgroup
2007  0::/
2008  # mkdir sub_cgrp_1
2009  # echo 0 > sub_cgrp_1/cgroup.procs
2010  # cat /proc/self/cgroup
2011  0::/sub_cgrp_1
2012
2013Each process gets its namespace-specific view of "/proc/$PID/cgroup"
2014
2015Processes running inside the cgroup namespace will be able to see
2016cgroup paths (in /proc/self/cgroup) only inside their root cgroup.
2017From within an unshared cgroupns::
2018
2019  # sleep 100000 &
2020  [1] 7353
2021  # echo 7353 > sub_cgrp_1/cgroup.procs
2022  # cat /proc/7353/cgroup
2023  0::/sub_cgrp_1
2024
2025From the initial cgroup namespace, the real cgroup path will be
2026visible::
2027
2028  $ cat /proc/7353/cgroup
2029  0::/batchjobs/container_id1/sub_cgrp_1
2030
2031From a sibling cgroup namespace (that is, a namespace rooted at a
2032different cgroup), the cgroup path relative to its own cgroup
2033namespace root will be shown.  For instance, if PID 7353's cgroup
2034namespace root is at '/batchjobs/container_id2', then it will see::
2035
2036  # cat /proc/7353/cgroup
2037  0::/../container_id2/sub_cgrp_1
2038
2039Note that the relative path always starts with '/' to indicate that
2040its relative to the cgroup namespace root of the caller.
2041
2042
2043Migration and setns(2)
2044----------------------
2045
2046Processes inside a cgroup namespace can move into and out of the
2047namespace root if they have proper access to external cgroups.  For
2048example, from inside a namespace with cgroupns root at
2049/batchjobs/container_id1, and assuming that the global hierarchy is
2050still accessible inside cgroupns::
2051
2052  # cat /proc/7353/cgroup
2053  0::/sub_cgrp_1
2054  # echo 7353 > batchjobs/container_id2/cgroup.procs
2055  # cat /proc/7353/cgroup
2056  0::/../container_id2
2057
2058Note that this kind of setup is not encouraged.  A task inside cgroup
2059namespace should only be exposed to its own cgroupns hierarchy.
2060
2061setns(2) to another cgroup namespace is allowed when:
2062
2063(a) the process has CAP_SYS_ADMIN against its current user namespace
2064(b) the process has CAP_SYS_ADMIN against the target cgroup
2065    namespace's userns
2066
2067No implicit cgroup changes happen with attaching to another cgroup
2068namespace.  It is expected that the someone moves the attaching
2069process under the target cgroup namespace root.
2070
2071
2072Interaction with Other Namespaces
2073---------------------------------
2074
2075Namespace specific cgroup hierarchy can be mounted by a process
2076running inside a non-init cgroup namespace::
2077
2078  # mount -t cgroup2 none $MOUNT_POINT
2079
2080This will mount the unified cgroup hierarchy with cgroupns root as the
2081filesystem root.  The process needs CAP_SYS_ADMIN against its user and
2082mount namespaces.
2083
2084The virtualization of /proc/self/cgroup file combined with restricting
2085the view of cgroup hierarchy by namespace-private cgroupfs mount
2086provides a properly isolated cgroup view inside the container.
2087
2088
2089Information on Kernel Programming
2090=================================
2091
2092This section contains kernel programming information in the areas
2093where interacting with cgroup is necessary.  cgroup core and
2094controllers are not covered.
2095
2096
2097Filesystem Support for Writeback
2098--------------------------------
2099
2100A filesystem can support cgroup writeback by updating
2101address_space_operations->writepage[s]() to annotate bio's using the
2102following two functions.
2103
2104  wbc_init_bio(@wbc, @bio)
2105	Should be called for each bio carrying writeback data and
2106	associates the bio with the inode's owner cgroup and the
2107	corresponding request queue.  This must be called after
2108	a queue (device) has been associated with the bio and
2109	before submission.
2110
2111  wbc_account_io(@wbc, @page, @bytes)
2112	Should be called for each data segment being written out.
2113	While this function doesn't care exactly when it's called
2114	during the writeback session, it's the easiest and most
2115	natural to call it as data segments are added to a bio.
2116
2117With writeback bio's annotated, cgroup support can be enabled per
2118super_block by setting SB_I_CGROUPWB in ->s_iflags.  This allows for
2119selective disabling of cgroup writeback support which is helpful when
2120certain filesystem features, e.g. journaled data mode, are
2121incompatible.
2122
2123wbc_init_bio() binds the specified bio to its cgroup.  Depending on
2124the configuration, the bio may be executed at a lower priority and if
2125the writeback session is holding shared resources, e.g. a journal
2126entry, may lead to priority inversion.  There is no one easy solution
2127for the problem.  Filesystems can try to work around specific problem
2128cases by skipping wbc_init_bio() and using bio_associate_blkg()
2129directly.
2130
2131
2132Deprecated v1 Core Features
2133===========================
2134
2135- Multiple hierarchies including named ones are not supported.
2136
2137- All v1 mount options are not supported.
2138
2139- The "tasks" file is removed and "cgroup.procs" is not sorted.
2140
2141- "cgroup.clone_children" is removed.
2142
2143- /proc/cgroups is meaningless for v2.  Use "cgroup.controllers" file
2144  at the root instead.
2145
2146
2147Issues with v1 and Rationales for v2
2148====================================
2149
2150Multiple Hierarchies
2151--------------------
2152
2153cgroup v1 allowed an arbitrary number of hierarchies and each
2154hierarchy could host any number of controllers.  While this seemed to
2155provide a high level of flexibility, it wasn't useful in practice.
2156
2157For example, as there is only one instance of each controller, utility
2158type controllers such as freezer which can be useful in all
2159hierarchies could only be used in one.  The issue is exacerbated by
2160the fact that controllers couldn't be moved to another hierarchy once
2161hierarchies were populated.  Another issue was that all controllers
2162bound to a hierarchy were forced to have exactly the same view of the
2163hierarchy.  It wasn't possible to vary the granularity depending on
2164the specific controller.
2165
2166In practice, these issues heavily limited which controllers could be
2167put on the same hierarchy and most configurations resorted to putting
2168each controller on its own hierarchy.  Only closely related ones, such
2169as the cpu and cpuacct controllers, made sense to be put on the same
2170hierarchy.  This often meant that userland ended up managing multiple
2171similar hierarchies repeating the same steps on each hierarchy
2172whenever a hierarchy management operation was necessary.
2173
2174Furthermore, support for multiple hierarchies came at a steep cost.
2175It greatly complicated cgroup core implementation but more importantly
2176the support for multiple hierarchies restricted how cgroup could be
2177used in general and what controllers was able to do.
2178
2179There was no limit on how many hierarchies there might be, which meant
2180that a thread's cgroup membership couldn't be described in finite
2181length.  The key might contain any number of entries and was unlimited
2182in length, which made it highly awkward to manipulate and led to
2183addition of controllers which existed only to identify membership,
2184which in turn exacerbated the original problem of proliferating number
2185of hierarchies.
2186
2187Also, as a controller couldn't have any expectation regarding the
2188topologies of hierarchies other controllers might be on, each
2189controller had to assume that all other controllers were attached to
2190completely orthogonal hierarchies.  This made it impossible, or at
2191least very cumbersome, for controllers to cooperate with each other.
2192
2193In most use cases, putting controllers on hierarchies which are
2194completely orthogonal to each other isn't necessary.  What usually is
2195called for is the ability to have differing levels of granularity
2196depending on the specific controller.  In other words, hierarchy may
2197be collapsed from leaf towards root when viewed from specific
2198controllers.  For example, a given configuration might not care about
2199how memory is distributed beyond a certain level while still wanting
2200to control how CPU cycles are distributed.
2201
2202
2203Thread Granularity
2204------------------
2205
2206cgroup v1 allowed threads of a process to belong to different cgroups.
2207This didn't make sense for some controllers and those controllers
2208ended up implementing different ways to ignore such situations but
2209much more importantly it blurred the line between API exposed to
2210individual applications and system management interface.
2211
2212Generally, in-process knowledge is available only to the process
2213itself; thus, unlike service-level organization of processes,
2214categorizing threads of a process requires active participation from
2215the application which owns the target process.
2216
2217cgroup v1 had an ambiguously defined delegation model which got abused
2218in combination with thread granularity.  cgroups were delegated to
2219individual applications so that they can create and manage their own
2220sub-hierarchies and control resource distributions along them.  This
2221effectively raised cgroup to the status of a syscall-like API exposed
2222to lay programs.
2223
2224First of all, cgroup has a fundamentally inadequate interface to be
2225exposed this way.  For a process to access its own knobs, it has to
2226extract the path on the target hierarchy from /proc/self/cgroup,
2227construct the path by appending the name of the knob to the path, open
2228and then read and/or write to it.  This is not only extremely clunky
2229and unusual but also inherently racy.  There is no conventional way to
2230define transaction across the required steps and nothing can guarantee
2231that the process would actually be operating on its own sub-hierarchy.
2232
2233cgroup controllers implemented a number of knobs which would never be
2234accepted as public APIs because they were just adding control knobs to
2235system-management pseudo filesystem.  cgroup ended up with interface
2236knobs which were not properly abstracted or refined and directly
2237revealed kernel internal details.  These knobs got exposed to
2238individual applications through the ill-defined delegation mechanism
2239effectively abusing cgroup as a shortcut to implementing public APIs
2240without going through the required scrutiny.
2241
2242This was painful for both userland and kernel.  Userland ended up with
2243misbehaving and poorly abstracted interfaces and kernel exposing and
2244locked into constructs inadvertently.
2245
2246
2247Competition Between Inner Nodes and Threads
2248-------------------------------------------
2249
2250cgroup v1 allowed threads to be in any cgroups which created an
2251interesting problem where threads belonging to a parent cgroup and its
2252children cgroups competed for resources.  This was nasty as two
2253different types of entities competed and there was no obvious way to
2254settle it.  Different controllers did different things.
2255
2256The cpu controller considered threads and cgroups as equivalents and
2257mapped nice levels to cgroup weights.  This worked for some cases but
2258fell flat when children wanted to be allocated specific ratios of CPU
2259cycles and the number of internal threads fluctuated - the ratios
2260constantly changed as the number of competing entities fluctuated.
2261There also were other issues.  The mapping from nice level to weight
2262wasn't obvious or universal, and there were various other knobs which
2263simply weren't available for threads.
2264
2265The io controller implicitly created a hidden leaf node for each
2266cgroup to host the threads.  The hidden leaf had its own copies of all
2267the knobs with ``leaf_`` prefixed.  While this allowed equivalent
2268control over internal threads, it was with serious drawbacks.  It
2269always added an extra layer of nesting which wouldn't be necessary
2270otherwise, made the interface messy and significantly complicated the
2271implementation.
2272
2273The memory controller didn't have a way to control what happened
2274between internal tasks and child cgroups and the behavior was not
2275clearly defined.  There were attempts to add ad-hoc behaviors and
2276knobs to tailor the behavior to specific workloads which would have
2277led to problems extremely difficult to resolve in the long term.
2278
2279Multiple controllers struggled with internal tasks and came up with
2280different ways to deal with it; unfortunately, all the approaches were
2281severely flawed and, furthermore, the widely different behaviors
2282made cgroup as a whole highly inconsistent.
2283
2284This clearly is a problem which needs to be addressed from cgroup core
2285in a uniform way.
2286
2287
2288Other Interface Issues
2289----------------------
2290
2291cgroup v1 grew without oversight and developed a large number of
2292idiosyncrasies and inconsistencies.  One issue on the cgroup core side
2293was how an empty cgroup was notified - a userland helper binary was
2294forked and executed for each event.  The event delivery wasn't
2295recursive or delegatable.  The limitations of the mechanism also led
2296to in-kernel event delivery filtering mechanism further complicating
2297the interface.
2298
2299Controller interfaces were problematic too.  An extreme example is
2300controllers completely ignoring hierarchical organization and treating
2301all cgroups as if they were all located directly under the root
2302cgroup.  Some controllers exposed a large amount of inconsistent
2303implementation details to userland.
2304
2305There also was no consistency across controllers.  When a new cgroup
2306was created, some controllers defaulted to not imposing extra
2307restrictions while others disallowed any resource usage until
2308explicitly configured.  Configuration knobs for the same type of
2309control used widely differing naming schemes and formats.  Statistics
2310and information knobs were named arbitrarily and used different
2311formats and units even in the same controller.
2312
2313cgroup v2 establishes common conventions where appropriate and updates
2314controllers so that they expose minimal and consistent interfaces.
2315
2316
2317Controller Issues and Remedies
2318------------------------------
2319
2320Memory
2321~~~~~~
2322
2323The original lower boundary, the soft limit, is defined as a limit
2324that is per default unset.  As a result, the set of cgroups that
2325global reclaim prefers is opt-in, rather than opt-out.  The costs for
2326optimizing these mostly negative lookups are so high that the
2327implementation, despite its enormous size, does not even provide the
2328basic desirable behavior.  First off, the soft limit has no
2329hierarchical meaning.  All configured groups are organized in a global
2330rbtree and treated like equal peers, regardless where they are located
2331in the hierarchy.  This makes subtree delegation impossible.  Second,
2332the soft limit reclaim pass is so aggressive that it not just
2333introduces high allocation latencies into the system, but also impacts
2334system performance due to overreclaim, to the point where the feature
2335becomes self-defeating.
2336
2337The memory.low boundary on the other hand is a top-down allocated
2338reserve.  A cgroup enjoys reclaim protection when it's within its low,
2339which makes delegation of subtrees possible.
2340
2341The original high boundary, the hard limit, is defined as a strict
2342limit that can not budge, even if the OOM killer has to be called.
2343But this generally goes against the goal of making the most out of the
2344available memory.  The memory consumption of workloads varies during
2345runtime, and that requires users to overcommit.  But doing that with a
2346strict upper limit requires either a fairly accurate prediction of the
2347working set size or adding slack to the limit.  Since working set size
2348estimation is hard and error prone, and getting it wrong results in
2349OOM kills, most users tend to err on the side of a looser limit and
2350end up wasting precious resources.
2351
2352The memory.high boundary on the other hand can be set much more
2353conservatively.  When hit, it throttles allocations by forcing them
2354into direct reclaim to work off the excess, but it never invokes the
2355OOM killer.  As a result, a high boundary that is chosen too
2356aggressively will not terminate the processes, but instead it will
2357lead to gradual performance degradation.  The user can monitor this
2358and make corrections until the minimal memory footprint that still
2359gives acceptable performance is found.
2360
2361In extreme cases, with many concurrent allocations and a complete
2362breakdown of reclaim progress within the group, the high boundary can
2363be exceeded.  But even then it's mostly better to satisfy the
2364allocation from the slack available in other groups or the rest of the
2365system than killing the group.  Otherwise, memory.max is there to
2366limit this type of spillover and ultimately contain buggy or even
2367malicious applications.
2368
2369Setting the original memory.limit_in_bytes below the current usage was
2370subject to a race condition, where concurrent charges could cause the
2371limit setting to fail. memory.max on the other hand will first set the
2372limit to prevent new charges, and then reclaim and OOM kill until the
2373new limit is met - or the task writing to memory.max is killed.
2374
2375The combined memory+swap accounting and limiting is replaced by real
2376control over swap space.
2377
2378The main argument for a combined memory+swap facility in the original
2379cgroup design was that global or parental pressure would always be
2380able to swap all anonymous memory of a child group, regardless of the
2381child's own (possibly untrusted) configuration.  However, untrusted
2382groups can sabotage swapping by other means - such as referencing its
2383anonymous memory in a tight loop - and an admin can not assume full
2384swappability when overcommitting untrusted jobs.
2385
2386For trusted jobs, on the other hand, a combined counter is not an
2387intuitive userspace interface, and it flies in the face of the idea
2388that cgroup controllers should account and limit specific physical
2389resources.  Swap space is a resource like all others in the system,
2390and that's why unified hierarchy allows distributing it separately.
2391