xref: /linux/Documentation/admin-guide/cgroup-v2.rst (revision f8e17c17b81070f38062dce79ca7f4541851dadd)
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/admin-guide/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. HugeTLB
65       5.8-1. HugeTLB Interface Files
66     5-8. Misc
67       5-8-1. perf_event
68     5-N. Non-normative information
69       5-N-1. CPU controller root cgroup process behaviour
70       5-N-2. IO controller root cgroup process behaviour
71   6. Namespace
72     6-1. Basics
73     6-2. The Root and Views
74     6-3. Migration and setns(2)
75     6-4. Interaction with Other Namespaces
76   P. Information on Kernel Programming
77     P-1. Filesystem Support for Writeback
78   D. Deprecated v1 Core Features
79   R. Issues with v1 and Rationales for v2
80     R-1. Multiple Hierarchies
81     R-2. Thread Granularity
82     R-3. Competition Between Inner Nodes and Threads
83     R-4. Other Interface Issues
84     R-5. Controller Issues and Remedies
85       R-5-1. Memory
86
87
88Introduction
89============
90
91Terminology
92-----------
93
94"cgroup" stands for "control group" and is never capitalized.  The
95singular form is used to designate the whole feature and also as a
96qualifier as in "cgroup controllers".  When explicitly referring to
97multiple individual control groups, the plural form "cgroups" is used.
98
99
100What is cgroup?
101---------------
102
103cgroup is a mechanism to organize processes hierarchically and
104distribute system resources along the hierarchy in a controlled and
105configurable manner.
106
107cgroup is largely composed of two parts - the core and controllers.
108cgroup core is primarily responsible for hierarchically organizing
109processes.  A cgroup controller is usually responsible for
110distributing a specific type of system resource along the hierarchy
111although there are utility controllers which serve purposes other than
112resource distribution.
113
114cgroups form a tree structure and every process in the system belongs
115to one and only one cgroup.  All threads of a process belong to the
116same cgroup.  On creation, all processes are put in the cgroup that
117the parent process belongs to at the time.  A process can be migrated
118to another cgroup.  Migration of a process doesn't affect already
119existing descendant processes.
120
121Following certain structural constraints, controllers may be enabled or
122disabled selectively on a cgroup.  All controller behaviors are
123hierarchical - if a controller is enabled on a cgroup, it affects all
124processes which belong to the cgroups consisting the inclusive
125sub-hierarchy of the cgroup.  When a controller is enabled on a nested
126cgroup, it always restricts the resource distribution further.  The
127restrictions set closer to the root in the hierarchy can not be
128overridden from further away.
129
130
131Basic Operations
132================
133
134Mounting
135--------
136
137Unlike v1, cgroup v2 has only single hierarchy.  The cgroup v2
138hierarchy can be mounted with the following mount command::
139
140  # mount -t cgroup2 none $MOUNT_POINT
141
142cgroup2 filesystem has the magic number 0x63677270 ("cgrp").  All
143controllers which support v2 and are not bound to a v1 hierarchy are
144automatically bound to the v2 hierarchy and show up at the root.
145Controllers which are not in active use in the v2 hierarchy can be
146bound to other hierarchies.  This allows mixing v2 hierarchy with the
147legacy v1 multiple hierarchies in a fully backward compatible way.
148
149A controller can be moved across hierarchies only after the controller
150is no longer referenced in its current hierarchy.  Because per-cgroup
151controller states are destroyed asynchronously and controllers may
152have lingering references, a controller may not show up immediately on
153the v2 hierarchy after the final umount of the previous hierarchy.
154Similarly, a controller should be fully disabled to be moved out of
155the unified hierarchy and it may take some time for the disabled
156controller to become available for other hierarchies; furthermore, due
157to inter-controller dependencies, other controllers may need to be
158disabled too.
159
160While useful for development and manual configurations, moving
161controllers dynamically between the v2 and other hierarchies is
162strongly discouraged for production use.  It is recommended to decide
163the hierarchies and controller associations before starting using the
164controllers after system boot.
165
166During transition to v2, system management software might still
167automount the v1 cgroup filesystem and so hijack all controllers
168during boot, before manual intervention is possible. To make testing
169and experimenting easier, the kernel parameter cgroup_no_v1= allows
170disabling controllers in v1 and make them always available in v2.
171
172cgroup v2 currently supports the following mount options.
173
174  nsdelegate
175
176	Consider cgroup namespaces as delegation boundaries.  This
177	option is system wide and can only be set on mount or modified
178	through remount from the init namespace.  The mount option is
179	ignored on non-init namespace mounts.  Please refer to the
180	Delegation section for details.
181
182  memory_localevents
183
184        Only populate memory.events with data for the current cgroup,
185        and not any subtrees. This is legacy behaviour, the default
186        behaviour without this option is to include subtree counts.
187        This option is system wide and can only be set on mount or
188        modified through remount from the init namespace. The mount
189        option is ignored on non-init namespace mounts.
190
191
192Organizing Processes and Threads
193--------------------------------
194
195Processes
196~~~~~~~~~
197
198Initially, only the root cgroup exists to which all processes belong.
199A child cgroup can be created by creating a sub-directory::
200
201  # mkdir $CGROUP_NAME
202
203A given cgroup may have multiple child cgroups forming a tree
204structure.  Each cgroup has a read-writable interface file
205"cgroup.procs".  When read, it lists the PIDs of all processes which
206belong to the cgroup one-per-line.  The PIDs are not ordered and the
207same PID may show up more than once if the process got moved to
208another cgroup and then back or the PID got recycled while reading.
209
210A process can be migrated into a cgroup by writing its PID to the
211target cgroup's "cgroup.procs" file.  Only one process can be migrated
212on a single write(2) call.  If a process is composed of multiple
213threads, writing the PID of any thread migrates all threads of the
214process.
215
216When a process forks a child process, the new process is born into the
217cgroup that the forking process belongs to at the time of the
218operation.  After exit, a process stays associated with the cgroup
219that it belonged to at the time of exit until it's reaped; however, a
220zombie process does not appear in "cgroup.procs" and thus can't be
221moved to another cgroup.
222
223A cgroup which doesn't have any children or live processes can be
224destroyed by removing the directory.  Note that a cgroup which doesn't
225have any children and is associated only with zombie processes is
226considered empty and can be removed::
227
228  # rmdir $CGROUP_NAME
229
230"/proc/$PID/cgroup" lists a process's cgroup membership.  If legacy
231cgroup is in use in the system, this file may contain multiple lines,
232one for each hierarchy.  The entry for cgroup v2 is always in the
233format "0::$PATH"::
234
235  # cat /proc/842/cgroup
236  ...
237  0::/test-cgroup/test-cgroup-nested
238
239If the process becomes a zombie and the cgroup it was associated with
240is removed subsequently, " (deleted)" is appended to the path::
241
242  # cat /proc/842/cgroup
243  ...
244  0::/test-cgroup/test-cgroup-nested (deleted)
245
246
247Threads
248~~~~~~~
249
250cgroup v2 supports thread granularity for a subset of controllers to
251support use cases requiring hierarchical resource distribution across
252the threads of a group of processes.  By default, all threads of a
253process belong to the same cgroup, which also serves as the resource
254domain to host resource consumptions which are not specific to a
255process or thread.  The thread mode allows threads to be spread across
256a subtree while still maintaining the common resource domain for them.
257
258Controllers which support thread mode are called threaded controllers.
259The ones which don't are called domain controllers.
260
261Marking a cgroup threaded makes it join the resource domain of its
262parent as a threaded cgroup.  The parent may be another threaded
263cgroup whose resource domain is further up in the hierarchy.  The root
264of a threaded subtree, that is, the nearest ancestor which is not
265threaded, is called threaded domain or thread root interchangeably and
266serves as the resource domain for the entire subtree.
267
268Inside a threaded subtree, threads of a process can be put in
269different cgroups and are not subject to the no internal process
270constraint - threaded controllers can be enabled on non-leaf cgroups
271whether they have threads in them or not.
272
273As the threaded domain cgroup hosts all the domain resource
274consumptions of the subtree, it is considered to have internal
275resource consumptions whether there are processes in it or not and
276can't have populated child cgroups which aren't threaded.  Because the
277root cgroup is not subject to no internal process constraint, it can
278serve both as a threaded domain and a parent to domain cgroups.
279
280The current operation mode or type of the cgroup is shown in the
281"cgroup.type" file which indicates whether the cgroup is a normal
282domain, a domain which is serving as the domain of a threaded subtree,
283or a threaded cgroup.
284
285On creation, a cgroup is always a domain cgroup and can be made
286threaded by writing "threaded" to the "cgroup.type" file.  The
287operation is single direction::
288
289  # echo threaded > cgroup.type
290
291Once threaded, the cgroup can't be made a domain again.  To enable the
292thread mode, the following conditions must be met.
293
294- As the cgroup will join the parent's resource domain.  The parent
295  must either be a valid (threaded) domain or a threaded cgroup.
296
297- When the parent is an unthreaded domain, it must not have any domain
298  controllers enabled or populated domain children.  The root is
299  exempt from this requirement.
300
301Topology-wise, a cgroup can be in an invalid state.  Please consider
302the following topology::
303
304  A (threaded domain) - B (threaded) - C (domain, just created)
305
306C is created as a domain but isn't connected to a parent which can
307host child domains.  C can't be used until it is turned into a
308threaded cgroup.  "cgroup.type" file will report "domain (invalid)" in
309these cases.  Operations which fail due to invalid topology use
310EOPNOTSUPP as the errno.
311
312A domain cgroup is turned into a threaded domain when one of its child
313cgroup becomes threaded or threaded controllers are enabled in the
314"cgroup.subtree_control" file while there are processes in the cgroup.
315A threaded domain reverts to a normal domain when the conditions
316clear.
317
318When read, "cgroup.threads" contains the list of the thread IDs of all
319threads in the cgroup.  Except that the operations are per-thread
320instead of per-process, "cgroup.threads" has the same format and
321behaves the same way as "cgroup.procs".  While "cgroup.threads" can be
322written to in any cgroup, as it can only move threads inside the same
323threaded domain, its operations are confined inside each threaded
324subtree.
325
326The threaded domain cgroup serves as the resource domain for the whole
327subtree, and, while the threads can be scattered across the subtree,
328all the processes are considered to be in the threaded domain cgroup.
329"cgroup.procs" in a threaded domain cgroup contains the PIDs of all
330processes in the subtree and is not readable in the subtree proper.
331However, "cgroup.procs" can be written to from anywhere in the subtree
332to migrate all threads of the matching process to the cgroup.
333
334Only threaded controllers can be enabled in a threaded subtree.  When
335a threaded controller is enabled inside a threaded subtree, it only
336accounts for and controls resource consumptions associated with the
337threads in the cgroup and its descendants.  All consumptions which
338aren't tied to a specific thread belong to the threaded domain cgroup.
339
340Because a threaded subtree is exempt from no internal process
341constraint, a threaded controller must be able to handle competition
342between threads in a non-leaf cgroup and its child cgroups.  Each
343threaded controller defines how such competitions are handled.
344
345
346[Un]populated Notification
347--------------------------
348
349Each non-root cgroup has a "cgroup.events" file which contains
350"populated" field indicating whether the cgroup's sub-hierarchy has
351live processes in it.  Its value is 0 if there is no live process in
352the cgroup and its descendants; otherwise, 1.  poll and [id]notify
353events are triggered when the value changes.  This can be used, for
354example, to start a clean-up operation after all processes of a given
355sub-hierarchy have exited.  The populated state updates and
356notifications are recursive.  Consider the following sub-hierarchy
357where the numbers in the parentheses represent the numbers of processes
358in each cgroup::
359
360  A(4) - B(0) - C(1)
361              \ D(0)
362
363A, B and C's "populated" fields would be 1 while D's 0.  After the one
364process in C exits, B and C's "populated" fields would flip to "0" and
365file modified events will be generated on the "cgroup.events" files of
366both cgroups.
367
368
369Controlling Controllers
370-----------------------
371
372Enabling and Disabling
373~~~~~~~~~~~~~~~~~~~~~~
374
375Each cgroup has a "cgroup.controllers" file which lists all
376controllers available for the cgroup to enable::
377
378  # cat cgroup.controllers
379  cpu io memory
380
381No controller is enabled by default.  Controllers can be enabled and
382disabled by writing to the "cgroup.subtree_control" file::
383
384  # echo "+cpu +memory -io" > cgroup.subtree_control
385
386Only controllers which are listed in "cgroup.controllers" can be
387enabled.  When multiple operations are specified as above, either they
388all succeed or fail.  If multiple operations on the same controller
389are specified, the last one is effective.
390
391Enabling a controller in a cgroup indicates that the distribution of
392the target resource across its immediate children will be controlled.
393Consider the following sub-hierarchy.  The enabled controllers are
394listed in parentheses::
395
396  A(cpu,memory) - B(memory) - C()
397                            \ D()
398
399As A has "cpu" and "memory" enabled, A will control the distribution
400of CPU cycles and memory to its children, in this case, B.  As B has
401"memory" enabled but not "CPU", C and D will compete freely on CPU
402cycles but their division of memory available to B will be controlled.
403
404As a controller regulates the distribution of the target resource to
405the cgroup's children, enabling it creates the controller's interface
406files in the child cgroups.  In the above example, enabling "cpu" on B
407would create the "cpu." prefixed controller interface files in C and
408D.  Likewise, disabling "memory" from B would remove the "memory."
409prefixed controller interface files from C and D.  This means that the
410controller interface files - anything which doesn't start with
411"cgroup." are owned by the parent rather than the cgroup itself.
412
413
414Top-down Constraint
415~~~~~~~~~~~~~~~~~~~
416
417Resources are distributed top-down and a cgroup can further distribute
418a resource only if the resource has been distributed to it from the
419parent.  This means that all non-root "cgroup.subtree_control" files
420can only contain controllers which are enabled in the parent's
421"cgroup.subtree_control" file.  A controller can be enabled only if
422the parent has the controller enabled and a controller can't be
423disabled if one or more children have it enabled.
424
425
426No Internal Process Constraint
427~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
428
429Non-root cgroups can distribute domain resources to their children
430only when they don't have any processes of their own.  In other words,
431only domain cgroups which don't contain any processes can have domain
432controllers enabled in their "cgroup.subtree_control" files.
433
434This guarantees that, when a domain controller is looking at the part
435of the hierarchy which has it enabled, processes are always only on
436the leaves.  This rules out situations where child cgroups compete
437against internal processes of the parent.
438
439The root cgroup is exempt from this restriction.  Root contains
440processes and anonymous resource consumption which can't be associated
441with any other cgroups and requires special treatment from most
442controllers.  How resource consumption in the root cgroup is governed
443is up to each controller (for more information on this topic please
444refer to the Non-normative information section in the Controllers
445chapter).
446
447Note that the restriction doesn't get in the way if there is no
448enabled controller in the cgroup's "cgroup.subtree_control".  This is
449important as otherwise it wouldn't be possible to create children of a
450populated cgroup.  To control resource distribution of a cgroup, the
451cgroup must create children and transfer all its processes to the
452children before enabling controllers in its "cgroup.subtree_control"
453file.
454
455
456Delegation
457----------
458
459Model of Delegation
460~~~~~~~~~~~~~~~~~~~
461
462A cgroup can be delegated in two ways.  First, to a less privileged
463user by granting write access of the directory and its "cgroup.procs",
464"cgroup.threads" and "cgroup.subtree_control" files to the user.
465Second, if the "nsdelegate" mount option is set, automatically to a
466cgroup namespace on namespace creation.
467
468Because the resource control interface files in a given directory
469control the distribution of the parent's resources, the delegatee
470shouldn't be allowed to write to them.  For the first method, this is
471achieved by not granting access to these files.  For the second, the
472kernel rejects writes to all files other than "cgroup.procs" and
473"cgroup.subtree_control" on a namespace root from inside the
474namespace.
475
476The end results are equivalent for both delegation types.  Once
477delegated, the user can build sub-hierarchy under the directory,
478organize processes inside it as it sees fit and further distribute the
479resources it received from the parent.  The limits and other settings
480of all resource controllers are hierarchical and regardless of what
481happens in the delegated sub-hierarchy, nothing can escape the
482resource restrictions imposed by the parent.
483
484Currently, cgroup doesn't impose any restrictions on the number of
485cgroups in or nesting depth of a delegated sub-hierarchy; however,
486this may be limited explicitly in the future.
487
488
489Delegation Containment
490~~~~~~~~~~~~~~~~~~~~~~
491
492A delegated sub-hierarchy is contained in the sense that processes
493can't be moved into or out of the sub-hierarchy by the delegatee.
494
495For delegations to a less privileged user, this is achieved by
496requiring the following conditions for a process with a non-root euid
497to migrate a target process into a cgroup by writing its PID to the
498"cgroup.procs" file.
499
500- The writer must have write access to the "cgroup.procs" file.
501
502- The writer must have write access to the "cgroup.procs" file of the
503  common ancestor of the source and destination cgroups.
504
505The above two constraints ensure that while a delegatee may migrate
506processes around freely in the delegated sub-hierarchy it can't pull
507in from or push out to outside the sub-hierarchy.
508
509For an example, let's assume cgroups C0 and C1 have been delegated to
510user U0 who created C00, C01 under C0 and C10 under C1 as follows and
511all processes under C0 and C1 belong to U0::
512
513  ~~~~~~~~~~~~~ - C0 - C00
514  ~ cgroup    ~      \ C01
515  ~ hierarchy ~
516  ~~~~~~~~~~~~~ - C1 - C10
517
518Let's also say U0 wants to write the PID of a process which is
519currently in C10 into "C00/cgroup.procs".  U0 has write access to the
520file; however, the common ancestor of the source cgroup C10 and the
521destination cgroup C00 is above the points of delegation and U0 would
522not have write access to its "cgroup.procs" files and thus the write
523will be denied with -EACCES.
524
525For delegations to namespaces, containment is achieved by requiring
526that both the source and destination cgroups are reachable from the
527namespace of the process which is attempting the migration.  If either
528is not reachable, the migration is rejected with -ENOENT.
529
530
531Guidelines
532----------
533
534Organize Once and Control
535~~~~~~~~~~~~~~~~~~~~~~~~~
536
537Migrating a process across cgroups is a relatively expensive operation
538and stateful resources such as memory are not moved together with the
539process.  This is an explicit design decision as there often exist
540inherent trade-offs between migration and various hot paths in terms
541of synchronization cost.
542
543As such, migrating processes across cgroups frequently as a means to
544apply different resource restrictions is discouraged.  A workload
545should be assigned to a cgroup according to the system's logical and
546resource structure once on start-up.  Dynamic adjustments to resource
547distribution can be made by changing controller configuration through
548the interface files.
549
550
551Avoid Name Collisions
552~~~~~~~~~~~~~~~~~~~~~
553
554Interface files for a cgroup and its children cgroups occupy the same
555directory and it is possible to create children cgroups which collide
556with interface files.
557
558All cgroup core interface files are prefixed with "cgroup." and each
559controller's interface files are prefixed with the controller name and
560a dot.  A controller's name is composed of lower case alphabets and
561'_'s but never begins with an '_' so it can be used as the prefix
562character for collision avoidance.  Also, interface file names won't
563start or end with terms which are often used in categorizing workloads
564such as job, service, slice, unit or workload.
565
566cgroup doesn't do anything to prevent name collisions and it's the
567user's responsibility to avoid them.
568
569
570Resource Distribution Models
571============================
572
573cgroup controllers implement several resource distribution schemes
574depending on the resource type and expected use cases.  This section
575describes major schemes in use along with their expected behaviors.
576
577
578Weights
579-------
580
581A parent's resource is distributed by adding up the weights of all
582active children and giving each the fraction matching the ratio of its
583weight against the sum.  As only children which can make use of the
584resource at the moment participate in the distribution, this is
585work-conserving.  Due to the dynamic nature, this model is usually
586used for stateless resources.
587
588All weights are in the range [1, 10000] with the default at 100.  This
589allows symmetric multiplicative biases in both directions at fine
590enough granularity while staying in the intuitive range.
591
592As long as the weight is in range, all configuration combinations are
593valid and there is no reason to reject configuration changes or
594process migrations.
595
596"cpu.weight" proportionally distributes CPU cycles to active children
597and is an example of this type.
598
599
600Limits
601------
602
603A child can only consume upto the configured amount of the resource.
604Limits can be over-committed - the sum of the limits of children can
605exceed the amount of resource available to the parent.
606
607Limits are in the range [0, max] and defaults to "max", which is noop.
608
609As limits can be over-committed, all configuration combinations are
610valid and there is no reason to reject configuration changes or
611process migrations.
612
613"io.max" limits the maximum BPS and/or IOPS that a cgroup can consume
614on an IO device and is an example of this type.
615
616
617Protections
618-----------
619
620A cgroup is protected upto the configured amount of the resource
621as long as the usages of all its ancestors are under their
622protected levels.  Protections can be hard guarantees or best effort
623soft boundaries.  Protections can also be over-committed in which case
624only upto the amount available to the parent is protected among
625children.
626
627Protections are in the range [0, max] and defaults to 0, which is
628noop.
629
630As protections can be over-committed, all configuration combinations
631are valid and there is no reason to reject configuration changes or
632process migrations.
633
634"memory.low" implements best-effort memory protection and is an
635example of this type.
636
637
638Allocations
639-----------
640
641A cgroup is exclusively allocated a certain amount of a finite
642resource.  Allocations can't be over-committed - the sum of the
643allocations of children can not exceed the amount of resource
644available to the parent.
645
646Allocations are in the range [0, max] and defaults to 0, which is no
647resource.
648
649As allocations can't be over-committed, some configuration
650combinations are invalid and should be rejected.  Also, if the
651resource is mandatory for execution of processes, process migrations
652may be rejected.
653
654"cpu.rt.max" hard-allocates realtime slices and is an example of this
655type.
656
657
658Interface Files
659===============
660
661Format
662------
663
664All interface files should be in one of the following formats whenever
665possible::
666
667  New-line separated values
668  (when only one value can be written at once)
669
670	VAL0\n
671	VAL1\n
672	...
673
674  Space separated values
675  (when read-only or multiple values can be written at once)
676
677	VAL0 VAL1 ...\n
678
679  Flat keyed
680
681	KEY0 VAL0\n
682	KEY1 VAL1\n
683	...
684
685  Nested keyed
686
687	KEY0 SUB_KEY0=VAL00 SUB_KEY1=VAL01...
688	KEY1 SUB_KEY0=VAL10 SUB_KEY1=VAL11...
689	...
690
691For a writable file, the format for writing should generally match
692reading; however, controllers may allow omitting later fields or
693implement restricted shortcuts for most common use cases.
694
695For both flat and nested keyed files, only the values for a single key
696can be written at a time.  For nested keyed files, the sub key pairs
697may be specified in any order and not all pairs have to be specified.
698
699
700Conventions
701-----------
702
703- Settings for a single feature should be contained in a single file.
704
705- The root cgroup should be exempt from resource control and thus
706  shouldn't have resource control interface files.  Also,
707  informational files on the root cgroup which end up showing global
708  information available elsewhere shouldn't exist.
709
710- The default time unit is microseconds.  If a different unit is ever
711  used, an explicit unit suffix must be present.
712
713- A parts-per quantity should use a percentage decimal with at least
714  two digit fractional part - e.g. 13.40.
715
716- If a controller implements weight based resource distribution, its
717  interface file should be named "weight" and have the range [1,
718  10000] with 100 as the default.  The values are chosen to allow
719  enough and symmetric bias in both directions while keeping it
720  intuitive (the default is 100%).
721
722- If a controller implements an absolute resource guarantee and/or
723  limit, the interface files should be named "min" and "max"
724  respectively.  If a controller implements best effort resource
725  guarantee and/or limit, the interface files should be named "low"
726  and "high" respectively.
727
728  In the above four control files, the special token "max" should be
729  used to represent upward infinity for both reading and writing.
730
731- If a setting has a configurable default value and keyed specific
732  overrides, the default entry should be keyed with "default" and
733  appear as the first entry in the file.
734
735  The default value can be updated by writing either "default $VAL" or
736  "$VAL".
737
738  When writing to update a specific override, "default" can be used as
739  the value to indicate removal of the override.  Override entries
740  with "default" as the value must not appear when read.
741
742  For example, a setting which is keyed by major:minor device numbers
743  with integer values may look like the following::
744
745    # cat cgroup-example-interface-file
746    default 150
747    8:0 300
748
749  The default value can be updated by::
750
751    # echo 125 > cgroup-example-interface-file
752
753  or::
754
755    # echo "default 125" > cgroup-example-interface-file
756
757  An override can be set by::
758
759    # echo "8:16 170" > cgroup-example-interface-file
760
761  and cleared by::
762
763    # echo "8:0 default" > cgroup-example-interface-file
764    # cat cgroup-example-interface-file
765    default 125
766    8:16 170
767
768- For events which are not very high frequency, an interface file
769  "events" should be created which lists event key value pairs.
770  Whenever a notifiable event happens, file modified event should be
771  generated on the file.
772
773
774Core Interface Files
775--------------------
776
777All cgroup core files are prefixed with "cgroup."
778
779  cgroup.type
780
781	A read-write single value file which exists on non-root
782	cgroups.
783
784	When read, it indicates the current type of the cgroup, which
785	can be one of the following values.
786
787	- "domain" : A normal valid domain cgroup.
788
789	- "domain threaded" : A threaded domain cgroup which is
790          serving as the root of a threaded subtree.
791
792	- "domain invalid" : A cgroup which is in an invalid state.
793	  It can't be populated or have controllers enabled.  It may
794	  be allowed to become a threaded cgroup.
795
796	- "threaded" : A threaded cgroup which is a member of a
797          threaded subtree.
798
799	A cgroup can be turned into a threaded cgroup by writing
800	"threaded" to this file.
801
802  cgroup.procs
803	A read-write new-line separated values file which exists on
804	all cgroups.
805
806	When read, it lists the PIDs of all processes which belong to
807	the cgroup one-per-line.  The PIDs are not ordered and the
808	same PID may show up more than once if the process got moved
809	to another cgroup and then back or the PID got recycled while
810	reading.
811
812	A PID can be written to migrate the process associated with
813	the PID to the cgroup.  The writer should match all of the
814	following conditions.
815
816	- It must have write access to the "cgroup.procs" file.
817
818	- It must have write access to the "cgroup.procs" file of the
819	  common ancestor of the source and destination cgroups.
820
821	When delegating a sub-hierarchy, write access to this file
822	should be granted along with the containing directory.
823
824	In a threaded cgroup, reading this file fails with EOPNOTSUPP
825	as all the processes belong to the thread root.  Writing is
826	supported and moves every thread of the process to the cgroup.
827
828  cgroup.threads
829	A read-write new-line separated values file which exists on
830	all cgroups.
831
832	When read, it lists the TIDs of all threads which belong to
833	the cgroup one-per-line.  The TIDs are not ordered and the
834	same TID may show up more than once if the thread got moved to
835	another cgroup and then back or the TID got recycled while
836	reading.
837
838	A TID can be written to migrate the thread associated with the
839	TID to the cgroup.  The writer should match all of the
840	following conditions.
841
842	- It must have write access to the "cgroup.threads" file.
843
844	- The cgroup that the thread is currently in must be in the
845          same resource domain as the destination cgroup.
846
847	- It must have write access to the "cgroup.procs" file of the
848	  common ancestor of the source and destination cgroups.
849
850	When delegating a sub-hierarchy, write access to this file
851	should be granted along with the containing directory.
852
853  cgroup.controllers
854	A read-only space separated values file which exists on all
855	cgroups.
856
857	It shows space separated list of all controllers available to
858	the cgroup.  The controllers are not ordered.
859
860  cgroup.subtree_control
861	A read-write space separated values file which exists on all
862	cgroups.  Starts out empty.
863
864	When read, it shows space separated list of the controllers
865	which are enabled to control resource distribution from the
866	cgroup to its children.
867
868	Space separated list of controllers prefixed with '+' or '-'
869	can be written to enable or disable controllers.  A controller
870	name prefixed with '+' enables the controller and '-'
871	disables.  If a controller appears more than once on the list,
872	the last one is effective.  When multiple enable and disable
873	operations are specified, either all succeed or all fail.
874
875  cgroup.events
876	A read-only flat-keyed file which exists on non-root cgroups.
877	The following entries are defined.  Unless specified
878	otherwise, a value change in this file generates a file
879	modified event.
880
881	  populated
882		1 if the cgroup or its descendants contains any live
883		processes; otherwise, 0.
884	  frozen
885		1 if the cgroup is frozen; otherwise, 0.
886
887  cgroup.max.descendants
888	A read-write single value files.  The default is "max".
889
890	Maximum allowed number of descent cgroups.
891	If the actual number of descendants is equal or larger,
892	an attempt to create a new cgroup in the hierarchy will fail.
893
894  cgroup.max.depth
895	A read-write single value files.  The default is "max".
896
897	Maximum allowed descent depth below the current cgroup.
898	If the actual descent depth is equal or larger,
899	an attempt to create a new child cgroup will fail.
900
901  cgroup.stat
902	A read-only flat-keyed file with the following entries:
903
904	  nr_descendants
905		Total number of visible descendant cgroups.
906
907	  nr_dying_descendants
908		Total number of dying descendant cgroups. A cgroup becomes
909		dying after being deleted by a user. The cgroup will remain
910		in dying state for some time undefined time (which can depend
911		on system load) before being completely destroyed.
912
913		A process can't enter a dying cgroup under any circumstances,
914		a dying cgroup can't revive.
915
916		A dying cgroup can consume system resources not exceeding
917		limits, which were active at the moment of cgroup deletion.
918
919  cgroup.freeze
920	A read-write single value file which exists on non-root cgroups.
921	Allowed values are "0" and "1". The default is "0".
922
923	Writing "1" to the file causes freezing of the cgroup and all
924	descendant cgroups. This means that all belonging processes will
925	be stopped and will not run until the cgroup will be explicitly
926	unfrozen. Freezing of the cgroup may take some time; when this action
927	is completed, the "frozen" value in the cgroup.events control file
928	will be updated to "1" and the corresponding notification will be
929	issued.
930
931	A cgroup can be frozen either by its own settings, or by settings
932	of any ancestor cgroups. If any of ancestor cgroups is frozen, the
933	cgroup will remain frozen.
934
935	Processes in the frozen cgroup can be killed by a fatal signal.
936	They also can enter and leave a frozen cgroup: either by an explicit
937	move by a user, or if freezing of the cgroup races with fork().
938	If a process is moved to a frozen cgroup, it stops. If a process is
939	moved out of a frozen cgroup, it becomes running.
940
941	Frozen status of a cgroup doesn't affect any cgroup tree operations:
942	it's possible to delete a frozen (and empty) cgroup, as well as
943	create new sub-cgroups.
944
945Controllers
946===========
947
948CPU
949---
950
951The "cpu" controllers regulates distribution of CPU cycles.  This
952controller implements weight and absolute bandwidth limit models for
953normal scheduling policy and absolute bandwidth allocation model for
954realtime scheduling policy.
955
956In all the above models, cycles distribution is defined only on a temporal
957base and it does not account for the frequency at which tasks are executed.
958The (optional) utilization clamping support allows to hint the schedutil
959cpufreq governor about the minimum desired frequency which should always be
960provided by a CPU, as well as the maximum desired frequency, which should not
961be exceeded by a CPU.
962
963WARNING: cgroup2 doesn't yet support control of realtime processes and
964the cpu controller can only be enabled when all RT processes are in
965the root cgroup.  Be aware that system management software may already
966have placed RT processes into nonroot cgroups during the system boot
967process, and these processes may need to be moved to the root cgroup
968before the cpu controller can be enabled.
969
970
971CPU Interface Files
972~~~~~~~~~~~~~~~~~~~
973
974All time durations are in microseconds.
975
976  cpu.stat
977	A read-only flat-keyed file which exists on non-root cgroups.
978	This file exists whether the controller is enabled or not.
979
980	It always reports the following three stats:
981
982	- usage_usec
983	- user_usec
984	- system_usec
985
986	and the following three when the controller is enabled:
987
988	- nr_periods
989	- nr_throttled
990	- throttled_usec
991
992  cpu.weight
993	A read-write single value file which exists on non-root
994	cgroups.  The default is "100".
995
996	The weight in the range [1, 10000].
997
998  cpu.weight.nice
999	A read-write single value file which exists on non-root
1000	cgroups.  The default is "0".
1001
1002	The nice value is in the range [-20, 19].
1003
1004	This interface file is an alternative interface for
1005	"cpu.weight" and allows reading and setting weight using the
1006	same values used by nice(2).  Because the range is smaller and
1007	granularity is coarser for the nice values, the read value is
1008	the closest approximation of the current weight.
1009
1010  cpu.max
1011	A read-write two value file which exists on non-root cgroups.
1012	The default is "max 100000".
1013
1014	The maximum bandwidth limit.  It's in the following format::
1015
1016	  $MAX $PERIOD
1017
1018	which indicates that the group may consume upto $MAX in each
1019	$PERIOD duration.  "max" for $MAX indicates no limit.  If only
1020	one number is written, $MAX is updated.
1021
1022  cpu.pressure
1023	A read-only nested-key file which exists on non-root cgroups.
1024
1025	Shows pressure stall information for CPU. See
1026	Documentation/accounting/psi.rst for details.
1027
1028  cpu.uclamp.min
1029        A read-write single value file which exists on non-root cgroups.
1030        The default is "0", i.e. no utilization boosting.
1031
1032        The requested minimum utilization (protection) as a percentage
1033        rational number, e.g. 12.34 for 12.34%.
1034
1035        This interface allows reading and setting minimum utilization clamp
1036        values similar to the sched_setattr(2). This minimum utilization
1037        value is used to clamp the task specific minimum utilization clamp.
1038
1039        The requested minimum utilization (protection) is always capped by
1040        the current value for the maximum utilization (limit), i.e.
1041        `cpu.uclamp.max`.
1042
1043  cpu.uclamp.max
1044        A read-write single value file which exists on non-root cgroups.
1045        The default is "max". i.e. no utilization capping
1046
1047        The requested maximum utilization (limit) as a percentage rational
1048        number, e.g. 98.76 for 98.76%.
1049
1050        This interface allows reading and setting maximum utilization clamp
1051        values similar to the sched_setattr(2). This maximum utilization
1052        value is used to clamp the task specific maximum utilization clamp.
1053
1054
1055
1056Memory
1057------
1058
1059The "memory" controller regulates distribution of memory.  Memory is
1060stateful and implements both limit and protection models.  Due to the
1061intertwining between memory usage and reclaim pressure and the
1062stateful nature of memory, the distribution model is relatively
1063complex.
1064
1065While not completely water-tight, all major memory usages by a given
1066cgroup are tracked so that the total memory consumption can be
1067accounted and controlled to a reasonable extent.  Currently, the
1068following types of memory usages are tracked.
1069
1070- Userland memory - page cache and anonymous memory.
1071
1072- Kernel data structures such as dentries and inodes.
1073
1074- TCP socket buffers.
1075
1076The above list may expand in the future for better coverage.
1077
1078
1079Memory Interface Files
1080~~~~~~~~~~~~~~~~~~~~~~
1081
1082All memory amounts are in bytes.  If a value which is not aligned to
1083PAGE_SIZE is written, the value may be rounded up to the closest
1084PAGE_SIZE multiple when read back.
1085
1086  memory.current
1087	A read-only single value file which exists on non-root
1088	cgroups.
1089
1090	The total amount of memory currently being used by the cgroup
1091	and its descendants.
1092
1093  memory.min
1094	A read-write single value file which exists on non-root
1095	cgroups.  The default is "0".
1096
1097	Hard memory protection.  If the memory usage of a cgroup
1098	is within its effective min boundary, the cgroup's memory
1099	won't be reclaimed under any conditions. If there is no
1100	unprotected reclaimable memory available, OOM killer
1101	is invoked. Above the effective min boundary (or
1102	effective low boundary if it is higher), pages are reclaimed
1103	proportionally to the overage, reducing reclaim pressure for
1104	smaller overages.
1105
1106       Effective min boundary is limited by memory.min values of
1107	all ancestor cgroups. If there is memory.min overcommitment
1108	(child cgroup or cgroups are requiring more protected memory
1109	than parent will allow), then each child cgroup will get
1110	the part of parent's protection proportional to its
1111	actual memory usage below memory.min.
1112
1113	Putting more memory than generally available under this
1114	protection is discouraged and may lead to constant OOMs.
1115
1116	If a memory cgroup is not populated with processes,
1117	its memory.min is ignored.
1118
1119  memory.low
1120	A read-write single value file which exists on non-root
1121	cgroups.  The default is "0".
1122
1123	Best-effort memory protection.  If the memory usage of a
1124	cgroup is within its effective low boundary, the cgroup's
1125	memory won't be reclaimed unless there is no reclaimable
1126	memory available in unprotected cgroups.
1127	Above the effective low	boundary (or
1128	effective min boundary if it is higher), pages are reclaimed
1129	proportionally to the overage, reducing reclaim pressure for
1130	smaller overages.
1131
1132	Effective low boundary is limited by memory.low values of
1133	all ancestor cgroups. If there is memory.low overcommitment
1134	(child cgroup or cgroups are requiring more protected memory
1135	than parent will allow), then each child cgroup will get
1136	the part of parent's protection proportional to its
1137	actual memory usage below memory.low.
1138
1139	Putting more memory than generally available under this
1140	protection is discouraged.
1141
1142  memory.high
1143	A read-write single value file which exists on non-root
1144	cgroups.  The default is "max".
1145
1146	Memory usage throttle limit.  This is the main mechanism to
1147	control memory usage of a cgroup.  If a cgroup's usage goes
1148	over the high boundary, the processes of the cgroup are
1149	throttled and put under heavy reclaim pressure.
1150
1151	Going over the high limit never invokes the OOM killer and
1152	under extreme conditions the limit may be breached.
1153
1154  memory.max
1155	A read-write single value file which exists on non-root
1156	cgroups.  The default is "max".
1157
1158	Memory usage hard limit.  This is the final protection
1159	mechanism.  If a cgroup's memory usage reaches this limit and
1160	can't be reduced, the OOM killer is invoked in the cgroup.
1161	Under certain circumstances, the usage may go over the limit
1162	temporarily.
1163
1164	This is the ultimate protection mechanism.  As long as the
1165	high limit is used and monitored properly, this limit's
1166	utility is limited to providing the final safety net.
1167
1168  memory.oom.group
1169	A read-write single value file which exists on non-root
1170	cgroups.  The default value is "0".
1171
1172	Determines whether the cgroup should be treated as
1173	an indivisible workload by the OOM killer. If set,
1174	all tasks belonging to the cgroup or to its descendants
1175	(if the memory cgroup is not a leaf cgroup) are killed
1176	together or not at all. This can be used to avoid
1177	partial kills to guarantee workload integrity.
1178
1179	Tasks with the OOM protection (oom_score_adj set to -1000)
1180	are treated as an exception and are never killed.
1181
1182	If the OOM killer is invoked in a cgroup, it's not going
1183	to kill any tasks outside of this cgroup, regardless
1184	memory.oom.group values of ancestor cgroups.
1185
1186  memory.events
1187	A read-only flat-keyed file which exists on non-root cgroups.
1188	The following entries are defined.  Unless specified
1189	otherwise, a value change in this file generates a file
1190	modified event.
1191
1192	Note that all fields in this file are hierarchical and the
1193	file modified event can be generated due to an event down the
1194	hierarchy. For for the local events at the cgroup level see
1195	memory.events.local.
1196
1197	  low
1198		The number of times the cgroup is reclaimed due to
1199		high memory pressure even though its usage is under
1200		the low boundary.  This usually indicates that the low
1201		boundary is over-committed.
1202
1203	  high
1204		The number of times processes of the cgroup are
1205		throttled and routed to perform direct memory reclaim
1206		because the high memory boundary was exceeded.  For a
1207		cgroup whose memory usage is capped by the high limit
1208		rather than global memory pressure, this event's
1209		occurrences are expected.
1210
1211	  max
1212		The number of times the cgroup's memory usage was
1213		about to go over the max boundary.  If direct reclaim
1214		fails to bring it down, the cgroup goes to OOM state.
1215
1216	  oom
1217		The number of time the cgroup's memory usage was
1218		reached the limit and allocation was about to fail.
1219
1220		Depending on context result could be invocation of OOM
1221		killer and retrying allocation or failing allocation.
1222
1223		Failed allocation in its turn could be returned into
1224		userspace as -ENOMEM or silently ignored in cases like
1225		disk readahead.  For now OOM in memory cgroup kills
1226		tasks iff shortage has happened inside page fault.
1227
1228		This event is not raised if the OOM killer is not
1229		considered as an option, e.g. for failed high-order
1230		allocations.
1231
1232	  oom_kill
1233		The number of processes belonging to this cgroup
1234		killed by any kind of OOM killer.
1235
1236  memory.events.local
1237	Similar to memory.events but the fields in the file are local
1238	to the cgroup i.e. not hierarchical. The file modified event
1239	generated on this file reflects only the local events.
1240
1241  memory.stat
1242	A read-only flat-keyed file which exists on non-root cgroups.
1243
1244	This breaks down the cgroup's memory footprint into different
1245	types of memory, type-specific details, and other information
1246	on the state and past events of the memory management system.
1247
1248	All memory amounts are in bytes.
1249
1250	The entries are ordered to be human readable, and new entries
1251	can show up in the middle. Don't rely on items remaining in a
1252	fixed position; use the keys to look up specific values!
1253
1254	  anon
1255		Amount of memory used in anonymous mappings such as
1256		brk(), sbrk(), and mmap(MAP_ANONYMOUS)
1257
1258	  file
1259		Amount of memory used to cache filesystem data,
1260		including tmpfs and shared memory.
1261
1262	  kernel_stack
1263		Amount of memory allocated to kernel stacks.
1264
1265	  slab
1266		Amount of memory used for storing in-kernel data
1267		structures.
1268
1269	  sock
1270		Amount of memory used in network transmission buffers
1271
1272	  shmem
1273		Amount of cached filesystem data that is swap-backed,
1274		such as tmpfs, shm segments, shared anonymous mmap()s
1275
1276	  file_mapped
1277		Amount of cached filesystem data mapped with mmap()
1278
1279	  file_dirty
1280		Amount of cached filesystem data that was modified but
1281		not yet written back to disk
1282
1283	  file_writeback
1284		Amount of cached filesystem data that was modified and
1285		is currently being written back to disk
1286
1287	  anon_thp
1288		Amount of memory used in anonymous mappings backed by
1289		transparent hugepages
1290
1291	  inactive_anon, active_anon, inactive_file, active_file, unevictable
1292		Amount of memory, swap-backed and filesystem-backed,
1293		on the internal memory management lists used by the
1294		page reclaim algorithm.
1295
1296		As these represent internal list state (eg. shmem pages are on anon
1297		memory management lists), inactive_foo + active_foo may not be equal to
1298		the value for the foo counter, since the foo counter is type-based, not
1299		list-based.
1300
1301	  slab_reclaimable
1302		Part of "slab" that might be reclaimed, such as
1303		dentries and inodes.
1304
1305	  slab_unreclaimable
1306		Part of "slab" that cannot be reclaimed on memory
1307		pressure.
1308
1309	  pgfault
1310		Total number of page faults incurred
1311
1312	  pgmajfault
1313		Number of major page faults incurred
1314
1315	  workingset_refault
1316
1317		Number of refaults of previously evicted pages
1318
1319	  workingset_activate
1320
1321		Number of refaulted pages that were immediately activated
1322
1323	  workingset_nodereclaim
1324
1325		Number of times a shadow node has been reclaimed
1326
1327	  pgrefill
1328
1329		Amount of scanned pages (in an active LRU list)
1330
1331	  pgscan
1332
1333		Amount of scanned pages (in an inactive LRU list)
1334
1335	  pgsteal
1336
1337		Amount of reclaimed pages
1338
1339	  pgactivate
1340
1341		Amount of pages moved to the active LRU list
1342
1343	  pgdeactivate
1344
1345		Amount of pages moved to the inactive LRU list
1346
1347	  pglazyfree
1348
1349		Amount of pages postponed to be freed under memory pressure
1350
1351	  pglazyfreed
1352
1353		Amount of reclaimed lazyfree pages
1354
1355	  thp_fault_alloc
1356
1357		Number of transparent hugepages which were allocated to satisfy
1358		a page fault, including COW faults. This counter is not present
1359		when CONFIG_TRANSPARENT_HUGEPAGE is not set.
1360
1361	  thp_collapse_alloc
1362
1363		Number of transparent hugepages which were allocated to allow
1364		collapsing an existing range of pages. This counter is not
1365		present when CONFIG_TRANSPARENT_HUGEPAGE is not set.
1366
1367  memory.swap.current
1368	A read-only single value file which exists on non-root
1369	cgroups.
1370
1371	The total amount of swap currently being used by the cgroup
1372	and its descendants.
1373
1374  memory.swap.max
1375	A read-write single value file which exists on non-root
1376	cgroups.  The default is "max".
1377
1378	Swap usage hard limit.  If a cgroup's swap usage reaches this
1379	limit, anonymous memory of the cgroup will not be swapped out.
1380
1381  memory.swap.events
1382	A read-only flat-keyed file which exists on non-root cgroups.
1383	The following entries are defined.  Unless specified
1384	otherwise, a value change in this file generates a file
1385	modified event.
1386
1387	  max
1388		The number of times the cgroup's swap usage was about
1389		to go over the max boundary and swap allocation
1390		failed.
1391
1392	  fail
1393		The number of times swap allocation failed either
1394		because of running out of swap system-wide or max
1395		limit.
1396
1397	When reduced under the current usage, the existing swap
1398	entries are reclaimed gradually and the swap usage may stay
1399	higher than the limit for an extended period of time.  This
1400	reduces the impact on the workload and memory management.
1401
1402  memory.pressure
1403	A read-only nested-key file which exists on non-root cgroups.
1404
1405	Shows pressure stall information for memory. See
1406	Documentation/accounting/psi.rst for details.
1407
1408
1409Usage Guidelines
1410~~~~~~~~~~~~~~~~
1411
1412"memory.high" is the main mechanism to control memory usage.
1413Over-committing on high limit (sum of high limits > available memory)
1414and letting global memory pressure to distribute memory according to
1415usage is a viable strategy.
1416
1417Because breach of the high limit doesn't trigger the OOM killer but
1418throttles the offending cgroup, a management agent has ample
1419opportunities to monitor and take appropriate actions such as granting
1420more memory or terminating the workload.
1421
1422Determining whether a cgroup has enough memory is not trivial as
1423memory usage doesn't indicate whether the workload can benefit from
1424more memory.  For example, a workload which writes data received from
1425network to a file can use all available memory but can also operate as
1426performant with a small amount of memory.  A measure of memory
1427pressure - how much the workload is being impacted due to lack of
1428memory - is necessary to determine whether a workload needs more
1429memory; unfortunately, memory pressure monitoring mechanism isn't
1430implemented yet.
1431
1432
1433Memory Ownership
1434~~~~~~~~~~~~~~~~
1435
1436A memory area is charged to the cgroup which instantiated it and stays
1437charged to the cgroup until the area is released.  Migrating a process
1438to a different cgroup doesn't move the memory usages that it
1439instantiated while in the previous cgroup to the new cgroup.
1440
1441A memory area may be used by processes belonging to different cgroups.
1442To which cgroup the area will be charged is in-deterministic; however,
1443over time, the memory area is likely to end up in a cgroup which has
1444enough memory allowance to avoid high reclaim pressure.
1445
1446If a cgroup sweeps a considerable amount of memory which is expected
1447to be accessed repeatedly by other cgroups, it may make sense to use
1448POSIX_FADV_DONTNEED to relinquish the ownership of memory areas
1449belonging to the affected files to ensure correct memory ownership.
1450
1451
1452IO
1453--
1454
1455The "io" controller regulates the distribution of IO resources.  This
1456controller implements both weight based and absolute bandwidth or IOPS
1457limit distribution; however, weight based distribution is available
1458only if cfq-iosched is in use and neither scheme is available for
1459blk-mq devices.
1460
1461
1462IO Interface Files
1463~~~~~~~~~~~~~~~~~~
1464
1465  io.stat
1466	A read-only nested-keyed file which exists on non-root
1467	cgroups.
1468
1469	Lines are keyed by $MAJ:$MIN device numbers and not ordered.
1470	The following nested keys are defined.
1471
1472	  ======	=====================
1473	  rbytes	Bytes read
1474	  wbytes	Bytes written
1475	  rios		Number of read IOs
1476	  wios		Number of write IOs
1477	  dbytes	Bytes discarded
1478	  dios		Number of discard IOs
1479	  ======	=====================
1480
1481	An example read output follows:
1482
1483	  8:16 rbytes=1459200 wbytes=314773504 rios=192 wios=353 dbytes=0 dios=0
1484	  8:0 rbytes=90430464 wbytes=299008000 rios=8950 wios=1252 dbytes=50331648 dios=3021
1485
1486  io.cost.qos
1487	A read-write nested-keyed file with exists only on the root
1488	cgroup.
1489
1490	This file configures the Quality of Service of the IO cost
1491	model based controller (CONFIG_BLK_CGROUP_IOCOST) which
1492	currently implements "io.weight" proportional control.  Lines
1493	are keyed by $MAJ:$MIN device numbers and not ordered.  The
1494	line for a given device is populated on the first write for
1495	the device on "io.cost.qos" or "io.cost.model".  The following
1496	nested keys are defined.
1497
1498	  ======	=====================================
1499	  enable	Weight-based control enable
1500	  ctrl		"auto" or "user"
1501	  rpct		Read latency percentile    [0, 100]
1502	  rlat		Read latency threshold
1503	  wpct		Write latency percentile   [0, 100]
1504	  wlat		Write latency threshold
1505	  min		Minimum scaling percentage [1, 10000]
1506	  max		Maximum scaling percentage [1, 10000]
1507	  ======	=====================================
1508
1509	The controller is disabled by default and can be enabled by
1510	setting "enable" to 1.  "rpct" and "wpct" parameters default
1511	to zero and the controller uses internal device saturation
1512	state to adjust the overall IO rate between "min" and "max".
1513
1514	When a better control quality is needed, latency QoS
1515	parameters can be configured.  For example::
1516
1517	  8:16 enable=1 ctrl=auto rpct=95.00 rlat=75000 wpct=95.00 wlat=150000 min=50.00 max=150.0
1518
1519	shows that on sdb, the controller is enabled, will consider
1520	the device saturated if the 95th percentile of read completion
1521	latencies is above 75ms or write 150ms, and adjust the overall
1522	IO issue rate between 50% and 150% accordingly.
1523
1524	The lower the saturation point, the better the latency QoS at
1525	the cost of aggregate bandwidth.  The narrower the allowed
1526	adjustment range between "min" and "max", the more conformant
1527	to the cost model the IO behavior.  Note that the IO issue
1528	base rate may be far off from 100% and setting "min" and "max"
1529	blindly can lead to a significant loss of device capacity or
1530	control quality.  "min" and "max" are useful for regulating
1531	devices which show wide temporary behavior changes - e.g. a
1532	ssd which accepts writes at the line speed for a while and
1533	then completely stalls for multiple seconds.
1534
1535	When "ctrl" is "auto", the parameters are controlled by the
1536	kernel and may change automatically.  Setting "ctrl" to "user"
1537	or setting any of the percentile and latency parameters puts
1538	it into "user" mode and disables the automatic changes.  The
1539	automatic mode can be restored by setting "ctrl" to "auto".
1540
1541  io.cost.model
1542	A read-write nested-keyed file with exists only on the root
1543	cgroup.
1544
1545	This file configures the cost model of the IO cost model based
1546	controller (CONFIG_BLK_CGROUP_IOCOST) which currently
1547	implements "io.weight" proportional control.  Lines are keyed
1548	by $MAJ:$MIN device numbers and not ordered.  The line for a
1549	given device is populated on the first write for the device on
1550	"io.cost.qos" or "io.cost.model".  The following nested keys
1551	are defined.
1552
1553	  =====		================================
1554	  ctrl		"auto" or "user"
1555	  model		The cost model in use - "linear"
1556	  =====		================================
1557
1558	When "ctrl" is "auto", the kernel may change all parameters
1559	dynamically.  When "ctrl" is set to "user" or any other
1560	parameters are written to, "ctrl" become "user" and the
1561	automatic changes are disabled.
1562
1563	When "model" is "linear", the following model parameters are
1564	defined.
1565
1566	  =============	========================================
1567	  [r|w]bps	The maximum sequential IO throughput
1568	  [r|w]seqiops	The maximum 4k sequential IOs per second
1569	  [r|w]randiops	The maximum 4k random IOs per second
1570	  =============	========================================
1571
1572	From the above, the builtin linear model determines the base
1573	costs of a sequential and random IO and the cost coefficient
1574	for the IO size.  While simple, this model can cover most
1575	common device classes acceptably.
1576
1577	The IO cost model isn't expected to be accurate in absolute
1578	sense and is scaled to the device behavior dynamically.
1579
1580	If needed, tools/cgroup/iocost_coef_gen.py can be used to
1581	generate device-specific coefficients.
1582
1583  io.weight
1584	A read-write flat-keyed file which exists on non-root cgroups.
1585	The default is "default 100".
1586
1587	The first line is the default weight applied to devices
1588	without specific override.  The rest are overrides keyed by
1589	$MAJ:$MIN device numbers and not ordered.  The weights are in
1590	the range [1, 10000] and specifies the relative amount IO time
1591	the cgroup can use in relation to its siblings.
1592
1593	The default weight can be updated by writing either "default
1594	$WEIGHT" or simply "$WEIGHT".  Overrides can be set by writing
1595	"$MAJ:$MIN $WEIGHT" and unset by writing "$MAJ:$MIN default".
1596
1597	An example read output follows::
1598
1599	  default 100
1600	  8:16 200
1601	  8:0 50
1602
1603  io.max
1604	A read-write nested-keyed file which exists on non-root
1605	cgroups.
1606
1607	BPS and IOPS based IO limit.  Lines are keyed by $MAJ:$MIN
1608	device numbers and not ordered.  The following nested keys are
1609	defined.
1610
1611	  =====		==================================
1612	  rbps		Max read bytes per second
1613	  wbps		Max write bytes per second
1614	  riops		Max read IO operations per second
1615	  wiops		Max write IO operations per second
1616	  =====		==================================
1617
1618	When writing, any number of nested key-value pairs can be
1619	specified in any order.  "max" can be specified as the value
1620	to remove a specific limit.  If the same key is specified
1621	multiple times, the outcome is undefined.
1622
1623	BPS and IOPS are measured in each IO direction and IOs are
1624	delayed if limit is reached.  Temporary bursts are allowed.
1625
1626	Setting read limit at 2M BPS and write at 120 IOPS for 8:16::
1627
1628	  echo "8:16 rbps=2097152 wiops=120" > io.max
1629
1630	Reading returns the following::
1631
1632	  8:16 rbps=2097152 wbps=max riops=max wiops=120
1633
1634	Write IOPS limit can be removed by writing the following::
1635
1636	  echo "8:16 wiops=max" > io.max
1637
1638	Reading now returns the following::
1639
1640	  8:16 rbps=2097152 wbps=max riops=max wiops=max
1641
1642  io.pressure
1643	A read-only nested-key file which exists on non-root cgroups.
1644
1645	Shows pressure stall information for IO. See
1646	Documentation/accounting/psi.rst for details.
1647
1648
1649Writeback
1650~~~~~~~~~
1651
1652Page cache is dirtied through buffered writes and shared mmaps and
1653written asynchronously to the backing filesystem by the writeback
1654mechanism.  Writeback sits between the memory and IO domains and
1655regulates the proportion of dirty memory by balancing dirtying and
1656write IOs.
1657
1658The io controller, in conjunction with the memory controller,
1659implements control of page cache writeback IOs.  The memory controller
1660defines the memory domain that dirty memory ratio is calculated and
1661maintained for and the io controller defines the io domain which
1662writes out dirty pages for the memory domain.  Both system-wide and
1663per-cgroup dirty memory states are examined and the more restrictive
1664of the two is enforced.
1665
1666cgroup writeback requires explicit support from the underlying
1667filesystem.  Currently, cgroup writeback is implemented on ext2, ext4
1668and btrfs.  On other filesystems, all writeback IOs are attributed to
1669the root cgroup.
1670
1671There are inherent differences in memory and writeback management
1672which affects how cgroup ownership is tracked.  Memory is tracked per
1673page while writeback per inode.  For the purpose of writeback, an
1674inode is assigned to a cgroup and all IO requests to write dirty pages
1675from the inode are attributed to that cgroup.
1676
1677As cgroup ownership for memory is tracked per page, there can be pages
1678which are associated with different cgroups than the one the inode is
1679associated with.  These are called foreign pages.  The writeback
1680constantly keeps track of foreign pages and, if a particular foreign
1681cgroup becomes the majority over a certain period of time, switches
1682the ownership of the inode to that cgroup.
1683
1684While this model is enough for most use cases where a given inode is
1685mostly dirtied by a single cgroup even when the main writing cgroup
1686changes over time, use cases where multiple cgroups write to a single
1687inode simultaneously are not supported well.  In such circumstances, a
1688significant portion of IOs are likely to be attributed incorrectly.
1689As memory controller assigns page ownership on the first use and
1690doesn't update it until the page is released, even if writeback
1691strictly follows page ownership, multiple cgroups dirtying overlapping
1692areas wouldn't work as expected.  It's recommended to avoid such usage
1693patterns.
1694
1695The sysctl knobs which affect writeback behavior are applied to cgroup
1696writeback as follows.
1697
1698  vm.dirty_background_ratio, vm.dirty_ratio
1699	These ratios apply the same to cgroup writeback with the
1700	amount of available memory capped by limits imposed by the
1701	memory controller and system-wide clean memory.
1702
1703  vm.dirty_background_bytes, vm.dirty_bytes
1704	For cgroup writeback, this is calculated into ratio against
1705	total available memory and applied the same way as
1706	vm.dirty[_background]_ratio.
1707
1708
1709IO Latency
1710~~~~~~~~~~
1711
1712This is a cgroup v2 controller for IO workload protection.  You provide a group
1713with a latency target, and if the average latency exceeds that target the
1714controller will throttle any peers that have a lower latency target than the
1715protected workload.
1716
1717The limits are only applied at the peer level in the hierarchy.  This means that
1718in the diagram below, only groups A, B, and C will influence each other, and
1719groups D and F will influence each other.  Group G will influence nobody::
1720
1721			[root]
1722		/	   |		\
1723		A	   B		C
1724	       /  \        |
1725	      D    F	   G
1726
1727
1728So the ideal way to configure this is to set io.latency in groups A, B, and C.
1729Generally you do not want to set a value lower than the latency your device
1730supports.  Experiment to find the value that works best for your workload.
1731Start at higher than the expected latency for your device and watch the
1732avg_lat value in io.stat for your workload group to get an idea of the
1733latency you see during normal operation.  Use the avg_lat value as a basis for
1734your real setting, setting at 10-15% higher than the value in io.stat.
1735
1736How IO Latency Throttling Works
1737~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1738
1739io.latency is work conserving; so as long as everybody is meeting their latency
1740target the controller doesn't do anything.  Once a group starts missing its
1741target it begins throttling any peer group that has a higher target than itself.
1742This throttling takes 2 forms:
1743
1744- Queue depth throttling.  This is the number of outstanding IO's a group is
1745  allowed to have.  We will clamp down relatively quickly, starting at no limit
1746  and going all the way down to 1 IO at a time.
1747
1748- Artificial delay induction.  There are certain types of IO that cannot be
1749  throttled without possibly adversely affecting higher priority groups.  This
1750  includes swapping and metadata IO.  These types of IO are allowed to occur
1751  normally, however they are "charged" to the originating group.  If the
1752  originating group is being throttled you will see the use_delay and delay
1753  fields in io.stat increase.  The delay value is how many microseconds that are
1754  being added to any process that runs in this group.  Because this number can
1755  grow quite large if there is a lot of swapping or metadata IO occurring we
1756  limit the individual delay events to 1 second at a time.
1757
1758Once the victimized group starts meeting its latency target again it will start
1759unthrottling any peer groups that were throttled previously.  If the victimized
1760group simply stops doing IO the global counter will unthrottle appropriately.
1761
1762IO Latency Interface Files
1763~~~~~~~~~~~~~~~~~~~~~~~~~~
1764
1765  io.latency
1766	This takes a similar format as the other controllers.
1767
1768		"MAJOR:MINOR target=<target time in microseconds"
1769
1770  io.stat
1771	If the controller is enabled you will see extra stats in io.stat in
1772	addition to the normal ones.
1773
1774	  depth
1775		This is the current queue depth for the group.
1776
1777	  avg_lat
1778		This is an exponential moving average with a decay rate of 1/exp
1779		bound by the sampling interval.  The decay rate interval can be
1780		calculated by multiplying the win value in io.stat by the
1781		corresponding number of samples based on the win value.
1782
1783	  win
1784		The sampling window size in milliseconds.  This is the minimum
1785		duration of time between evaluation events.  Windows only elapse
1786		with IO activity.  Idle periods extend the most recent window.
1787
1788PID
1789---
1790
1791The process number controller is used to allow a cgroup to stop any
1792new tasks from being fork()'d or clone()'d after a specified limit is
1793reached.
1794
1795The number of tasks in a cgroup can be exhausted in ways which other
1796controllers cannot prevent, thus warranting its own controller.  For
1797example, a fork bomb is likely to exhaust the number of tasks before
1798hitting memory restrictions.
1799
1800Note that PIDs used in this controller refer to TIDs, process IDs as
1801used by the kernel.
1802
1803
1804PID Interface Files
1805~~~~~~~~~~~~~~~~~~~
1806
1807  pids.max
1808	A read-write single value file which exists on non-root
1809	cgroups.  The default is "max".
1810
1811	Hard limit of number of processes.
1812
1813  pids.current
1814	A read-only single value file which exists on all cgroups.
1815
1816	The number of processes currently in the cgroup and its
1817	descendants.
1818
1819Organisational operations are not blocked by cgroup policies, so it is
1820possible to have pids.current > pids.max.  This can be done by either
1821setting the limit to be smaller than pids.current, or attaching enough
1822processes to the cgroup such that pids.current is larger than
1823pids.max.  However, it is not possible to violate a cgroup PID policy
1824through fork() or clone(). These will return -EAGAIN if the creation
1825of a new process would cause a cgroup policy to be violated.
1826
1827
1828Cpuset
1829------
1830
1831The "cpuset" controller provides a mechanism for constraining
1832the CPU and memory node placement of tasks to only the resources
1833specified in the cpuset interface files in a task's current cgroup.
1834This is especially valuable on large NUMA systems where placing jobs
1835on properly sized subsets of the systems with careful processor and
1836memory placement to reduce cross-node memory access and contention
1837can improve overall system performance.
1838
1839The "cpuset" controller is hierarchical.  That means the controller
1840cannot use CPUs or memory nodes not allowed in its parent.
1841
1842
1843Cpuset Interface Files
1844~~~~~~~~~~~~~~~~~~~~~~
1845
1846  cpuset.cpus
1847	A read-write multiple values file which exists on non-root
1848	cpuset-enabled cgroups.
1849
1850	It lists the requested CPUs to be used by tasks within this
1851	cgroup.  The actual list of CPUs to be granted, however, is
1852	subjected to constraints imposed by its parent and can differ
1853	from the requested CPUs.
1854
1855	The CPU numbers are comma-separated numbers or ranges.
1856	For example:
1857
1858	  # cat cpuset.cpus
1859	  0-4,6,8-10
1860
1861	An empty value indicates that the cgroup is using the same
1862	setting as the nearest cgroup ancestor with a non-empty
1863	"cpuset.cpus" or all the available CPUs if none is found.
1864
1865	The value of "cpuset.cpus" stays constant until the next update
1866	and won't be affected by any CPU hotplug events.
1867
1868  cpuset.cpus.effective
1869	A read-only multiple values file which exists on all
1870	cpuset-enabled cgroups.
1871
1872	It lists the onlined CPUs that are actually granted to this
1873	cgroup by its parent.  These CPUs are allowed to be used by
1874	tasks within the current cgroup.
1875
1876	If "cpuset.cpus" is empty, the "cpuset.cpus.effective" file shows
1877	all the CPUs from the parent cgroup that can be available to
1878	be used by this cgroup.  Otherwise, it should be a subset of
1879	"cpuset.cpus" unless none of the CPUs listed in "cpuset.cpus"
1880	can be granted.  In this case, it will be treated just like an
1881	empty "cpuset.cpus".
1882
1883	Its value will be affected by CPU hotplug events.
1884
1885  cpuset.mems
1886	A read-write multiple values file which exists on non-root
1887	cpuset-enabled cgroups.
1888
1889	It lists the requested memory nodes to be used by tasks within
1890	this cgroup.  The actual list of memory nodes granted, however,
1891	is subjected to constraints imposed by its parent and can differ
1892	from the requested memory nodes.
1893
1894	The memory node numbers are comma-separated numbers or ranges.
1895	For example:
1896
1897	  # cat cpuset.mems
1898	  0-1,3
1899
1900	An empty value indicates that the cgroup is using the same
1901	setting as the nearest cgroup ancestor with a non-empty
1902	"cpuset.mems" or all the available memory nodes if none
1903	is found.
1904
1905	The value of "cpuset.mems" stays constant until the next update
1906	and won't be affected by any memory nodes hotplug events.
1907
1908  cpuset.mems.effective
1909	A read-only multiple values file which exists on all
1910	cpuset-enabled cgroups.
1911
1912	It lists the onlined memory nodes that are actually granted to
1913	this cgroup by its parent. These memory nodes are allowed to
1914	be used by tasks within the current cgroup.
1915
1916	If "cpuset.mems" is empty, it shows all the memory nodes from the
1917	parent cgroup that will be available to be used by this cgroup.
1918	Otherwise, it should be a subset of "cpuset.mems" unless none of
1919	the memory nodes listed in "cpuset.mems" can be granted.  In this
1920	case, it will be treated just like an empty "cpuset.mems".
1921
1922	Its value will be affected by memory nodes hotplug events.
1923
1924  cpuset.cpus.partition
1925	A read-write single value file which exists on non-root
1926	cpuset-enabled cgroups.  This flag is owned by the parent cgroup
1927	and is not delegatable.
1928
1929        It accepts only the following input values when written to.
1930
1931        "root"   - a partition root
1932        "member" - a non-root member of a partition
1933
1934	When set to be a partition root, the current cgroup is the
1935	root of a new partition or scheduling domain that comprises
1936	itself and all its descendants except those that are separate
1937	partition roots themselves and their descendants.  The root
1938	cgroup is always a partition root.
1939
1940	There are constraints on where a partition root can be set.
1941	It can only be set in a cgroup if all the following conditions
1942	are true.
1943
1944	1) The "cpuset.cpus" is not empty and the list of CPUs are
1945	   exclusive, i.e. they are not shared by any of its siblings.
1946	2) The parent cgroup is a partition root.
1947	3) The "cpuset.cpus" is also a proper subset of the parent's
1948	   "cpuset.cpus.effective".
1949	4) There is no child cgroups with cpuset enabled.  This is for
1950	   eliminating corner cases that have to be handled if such a
1951	   condition is allowed.
1952
1953	Setting it to partition root will take the CPUs away from the
1954	effective CPUs of the parent cgroup.  Once it is set, this
1955	file cannot be reverted back to "member" if there are any child
1956	cgroups with cpuset enabled.
1957
1958	A parent partition cannot distribute all its CPUs to its
1959	child partitions.  There must be at least one cpu left in the
1960	parent partition.
1961
1962	Once becoming a partition root, changes to "cpuset.cpus" is
1963	generally allowed as long as the first condition above is true,
1964	the change will not take away all the CPUs from the parent
1965	partition and the new "cpuset.cpus" value is a superset of its
1966	children's "cpuset.cpus" values.
1967
1968	Sometimes, external factors like changes to ancestors'
1969	"cpuset.cpus" or cpu hotplug can cause the state of the partition
1970	root to change.  On read, the "cpuset.sched.partition" file
1971	can show the following values.
1972
1973	"member"       Non-root member of a partition
1974	"root"         Partition root
1975	"root invalid" Invalid partition root
1976
1977	It is a partition root if the first 2 partition root conditions
1978	above are true and at least one CPU from "cpuset.cpus" is
1979	granted by the parent cgroup.
1980
1981	A partition root can become invalid if none of CPUs requested
1982	in "cpuset.cpus" can be granted by the parent cgroup or the
1983	parent cgroup is no longer a partition root itself.  In this
1984	case, it is not a real partition even though the restriction
1985	of the first partition root condition above will still apply.
1986	The cpu affinity of all the tasks in the cgroup will then be
1987	associated with CPUs in the nearest ancestor partition.
1988
1989	An invalid partition root can be transitioned back to a
1990	real partition root if at least one of the requested CPUs
1991	can now be granted by its parent.  In this case, the cpu
1992	affinity of all the tasks in the formerly invalid partition
1993	will be associated to the CPUs of the newly formed partition.
1994	Changing the partition state of an invalid partition root to
1995	"member" is always allowed even if child cpusets are present.
1996
1997
1998Device controller
1999-----------------
2000
2001Device controller manages access to device files. It includes both
2002creation of new device files (using mknod), and access to the
2003existing device files.
2004
2005Cgroup v2 device controller has no interface files and is implemented
2006on top of cgroup BPF. To control access to device files, a user may
2007create bpf programs of the BPF_CGROUP_DEVICE type and attach them
2008to cgroups. On an attempt to access a device file, corresponding
2009BPF programs will be executed, and depending on the return value
2010the attempt will succeed or fail with -EPERM.
2011
2012A BPF_CGROUP_DEVICE program takes a pointer to the bpf_cgroup_dev_ctx
2013structure, which describes the device access attempt: access type
2014(mknod/read/write) and device (type, major and minor numbers).
2015If the program returns 0, the attempt fails with -EPERM, otherwise
2016it succeeds.
2017
2018An example of BPF_CGROUP_DEVICE program may be found in the kernel
2019source tree in the tools/testing/selftests/bpf/dev_cgroup.c file.
2020
2021
2022RDMA
2023----
2024
2025The "rdma" controller regulates the distribution and accounting of
2026of RDMA resources.
2027
2028RDMA Interface Files
2029~~~~~~~~~~~~~~~~~~~~
2030
2031  rdma.max
2032	A readwrite nested-keyed file that exists for all the cgroups
2033	except root that describes current configured resource limit
2034	for a RDMA/IB device.
2035
2036	Lines are keyed by device name and are not ordered.
2037	Each line contains space separated resource name and its configured
2038	limit that can be distributed.
2039
2040	The following nested keys are defined.
2041
2042	  ==========	=============================
2043	  hca_handle	Maximum number of HCA Handles
2044	  hca_object 	Maximum number of HCA Objects
2045	  ==========	=============================
2046
2047	An example for mlx4 and ocrdma device follows::
2048
2049	  mlx4_0 hca_handle=2 hca_object=2000
2050	  ocrdma1 hca_handle=3 hca_object=max
2051
2052  rdma.current
2053	A read-only file that describes current resource usage.
2054	It exists for all the cgroup except root.
2055
2056	An example for mlx4 and ocrdma device follows::
2057
2058	  mlx4_0 hca_handle=1 hca_object=20
2059	  ocrdma1 hca_handle=1 hca_object=23
2060
2061HugeTLB
2062-------
2063
2064The HugeTLB controller allows to limit the HugeTLB usage per control group and
2065enforces the controller limit during page fault.
2066
2067HugeTLB Interface Files
2068~~~~~~~~~~~~~~~~~~~~~~~
2069
2070  hugetlb.<hugepagesize>.current
2071	Show current usage for "hugepagesize" hugetlb.  It exists for all
2072	the cgroup except root.
2073
2074  hugetlb.<hugepagesize>.max
2075	Set/show the hard limit of "hugepagesize" hugetlb usage.
2076	The default value is "max".  It exists for all the cgroup except root.
2077
2078  hugetlb.<hugepagesize>.events
2079	A read-only flat-keyed file which exists on non-root cgroups.
2080
2081	  max
2082		The number of allocation failure due to HugeTLB limit
2083
2084  hugetlb.<hugepagesize>.events.local
2085	Similar to hugetlb.<hugepagesize>.events but the fields in the file
2086	are local to the cgroup i.e. not hierarchical. The file modified event
2087	generated on this file reflects only the local events.
2088
2089Misc
2090----
2091
2092perf_event
2093~~~~~~~~~~
2094
2095perf_event controller, if not mounted on a legacy hierarchy, is
2096automatically enabled on the v2 hierarchy so that perf events can
2097always be filtered by cgroup v2 path.  The controller can still be
2098moved to a legacy hierarchy after v2 hierarchy is populated.
2099
2100
2101Non-normative information
2102-------------------------
2103
2104This section contains information that isn't considered to be a part of
2105the stable kernel API and so is subject to change.
2106
2107
2108CPU controller root cgroup process behaviour
2109~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2110
2111When distributing CPU cycles in the root cgroup each thread in this
2112cgroup is treated as if it was hosted in a separate child cgroup of the
2113root cgroup. This child cgroup weight is dependent on its thread nice
2114level.
2115
2116For details of this mapping see sched_prio_to_weight array in
2117kernel/sched/core.c file (values from this array should be scaled
2118appropriately so the neutral - nice 0 - value is 100 instead of 1024).
2119
2120
2121IO controller root cgroup process behaviour
2122~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2123
2124Root cgroup processes are hosted in an implicit leaf child node.
2125When distributing IO resources this implicit child node is taken into
2126account as if it was a normal child cgroup of the root cgroup with a
2127weight value of 200.
2128
2129
2130Namespace
2131=========
2132
2133Basics
2134------
2135
2136cgroup namespace provides a mechanism to virtualize the view of the
2137"/proc/$PID/cgroup" file and cgroup mounts.  The CLONE_NEWCGROUP clone
2138flag can be used with clone(2) and unshare(2) to create a new cgroup
2139namespace.  The process running inside the cgroup namespace will have
2140its "/proc/$PID/cgroup" output restricted to cgroupns root.  The
2141cgroupns root is the cgroup of the process at the time of creation of
2142the cgroup namespace.
2143
2144Without cgroup namespace, the "/proc/$PID/cgroup" file shows the
2145complete path of the cgroup of a process.  In a container setup where
2146a set of cgroups and namespaces are intended to isolate processes the
2147"/proc/$PID/cgroup" file may leak potential system level information
2148to the isolated processes.  For Example::
2149
2150  # cat /proc/self/cgroup
2151  0::/batchjobs/container_id1
2152
2153The path '/batchjobs/container_id1' can be considered as system-data
2154and undesirable to expose to the isolated processes.  cgroup namespace
2155can be used to restrict visibility of this path.  For example, before
2156creating a cgroup namespace, one would see::
2157
2158  # ls -l /proc/self/ns/cgroup
2159  lrwxrwxrwx 1 root root 0 2014-07-15 10:37 /proc/self/ns/cgroup -> cgroup:[4026531835]
2160  # cat /proc/self/cgroup
2161  0::/batchjobs/container_id1
2162
2163After unsharing a new namespace, the view changes::
2164
2165  # ls -l /proc/self/ns/cgroup
2166  lrwxrwxrwx 1 root root 0 2014-07-15 10:35 /proc/self/ns/cgroup -> cgroup:[4026532183]
2167  # cat /proc/self/cgroup
2168  0::/
2169
2170When some thread from a multi-threaded process unshares its cgroup
2171namespace, the new cgroupns gets applied to the entire process (all
2172the threads).  This is natural for the v2 hierarchy; however, for the
2173legacy hierarchies, this may be unexpected.
2174
2175A cgroup namespace is alive as long as there are processes inside or
2176mounts pinning it.  When the last usage goes away, the cgroup
2177namespace is destroyed.  The cgroupns root and the actual cgroups
2178remain.
2179
2180
2181The Root and Views
2182------------------
2183
2184The 'cgroupns root' for a cgroup namespace is the cgroup in which the
2185process calling unshare(2) is running.  For example, if a process in
2186/batchjobs/container_id1 cgroup calls unshare, cgroup
2187/batchjobs/container_id1 becomes the cgroupns root.  For the
2188init_cgroup_ns, this is the real root ('/') cgroup.
2189
2190The cgroupns root cgroup does not change even if the namespace creator
2191process later moves to a different cgroup::
2192
2193  # ~/unshare -c # unshare cgroupns in some cgroup
2194  # cat /proc/self/cgroup
2195  0::/
2196  # mkdir sub_cgrp_1
2197  # echo 0 > sub_cgrp_1/cgroup.procs
2198  # cat /proc/self/cgroup
2199  0::/sub_cgrp_1
2200
2201Each process gets its namespace-specific view of "/proc/$PID/cgroup"
2202
2203Processes running inside the cgroup namespace will be able to see
2204cgroup paths (in /proc/self/cgroup) only inside their root cgroup.
2205From within an unshared cgroupns::
2206
2207  # sleep 100000 &
2208  [1] 7353
2209  # echo 7353 > sub_cgrp_1/cgroup.procs
2210  # cat /proc/7353/cgroup
2211  0::/sub_cgrp_1
2212
2213From the initial cgroup namespace, the real cgroup path will be
2214visible::
2215
2216  $ cat /proc/7353/cgroup
2217  0::/batchjobs/container_id1/sub_cgrp_1
2218
2219From a sibling cgroup namespace (that is, a namespace rooted at a
2220different cgroup), the cgroup path relative to its own cgroup
2221namespace root will be shown.  For instance, if PID 7353's cgroup
2222namespace root is at '/batchjobs/container_id2', then it will see::
2223
2224  # cat /proc/7353/cgroup
2225  0::/../container_id2/sub_cgrp_1
2226
2227Note that the relative path always starts with '/' to indicate that
2228its relative to the cgroup namespace root of the caller.
2229
2230
2231Migration and setns(2)
2232----------------------
2233
2234Processes inside a cgroup namespace can move into and out of the
2235namespace root if they have proper access to external cgroups.  For
2236example, from inside a namespace with cgroupns root at
2237/batchjobs/container_id1, and assuming that the global hierarchy is
2238still accessible inside cgroupns::
2239
2240  # cat /proc/7353/cgroup
2241  0::/sub_cgrp_1
2242  # echo 7353 > batchjobs/container_id2/cgroup.procs
2243  # cat /proc/7353/cgroup
2244  0::/../container_id2
2245
2246Note that this kind of setup is not encouraged.  A task inside cgroup
2247namespace should only be exposed to its own cgroupns hierarchy.
2248
2249setns(2) to another cgroup namespace is allowed when:
2250
2251(a) the process has CAP_SYS_ADMIN against its current user namespace
2252(b) the process has CAP_SYS_ADMIN against the target cgroup
2253    namespace's userns
2254
2255No implicit cgroup changes happen with attaching to another cgroup
2256namespace.  It is expected that the someone moves the attaching
2257process under the target cgroup namespace root.
2258
2259
2260Interaction with Other Namespaces
2261---------------------------------
2262
2263Namespace specific cgroup hierarchy can be mounted by a process
2264running inside a non-init cgroup namespace::
2265
2266  # mount -t cgroup2 none $MOUNT_POINT
2267
2268This will mount the unified cgroup hierarchy with cgroupns root as the
2269filesystem root.  The process needs CAP_SYS_ADMIN against its user and
2270mount namespaces.
2271
2272The virtualization of /proc/self/cgroup file combined with restricting
2273the view of cgroup hierarchy by namespace-private cgroupfs mount
2274provides a properly isolated cgroup view inside the container.
2275
2276
2277Information on Kernel Programming
2278=================================
2279
2280This section contains kernel programming information in the areas
2281where interacting with cgroup is necessary.  cgroup core and
2282controllers are not covered.
2283
2284
2285Filesystem Support for Writeback
2286--------------------------------
2287
2288A filesystem can support cgroup writeback by updating
2289address_space_operations->writepage[s]() to annotate bio's using the
2290following two functions.
2291
2292  wbc_init_bio(@wbc, @bio)
2293	Should be called for each bio carrying writeback data and
2294	associates the bio with the inode's owner cgroup and the
2295	corresponding request queue.  This must be called after
2296	a queue (device) has been associated with the bio and
2297	before submission.
2298
2299  wbc_account_cgroup_owner(@wbc, @page, @bytes)
2300	Should be called for each data segment being written out.
2301	While this function doesn't care exactly when it's called
2302	during the writeback session, it's the easiest and most
2303	natural to call it as data segments are added to a bio.
2304
2305With writeback bio's annotated, cgroup support can be enabled per
2306super_block by setting SB_I_CGROUPWB in ->s_iflags.  This allows for
2307selective disabling of cgroup writeback support which is helpful when
2308certain filesystem features, e.g. journaled data mode, are
2309incompatible.
2310
2311wbc_init_bio() binds the specified bio to its cgroup.  Depending on
2312the configuration, the bio may be executed at a lower priority and if
2313the writeback session is holding shared resources, e.g. a journal
2314entry, may lead to priority inversion.  There is no one easy solution
2315for the problem.  Filesystems can try to work around specific problem
2316cases by skipping wbc_init_bio() and using bio_associate_blkg()
2317directly.
2318
2319
2320Deprecated v1 Core Features
2321===========================
2322
2323- Multiple hierarchies including named ones are not supported.
2324
2325- All v1 mount options are not supported.
2326
2327- The "tasks" file is removed and "cgroup.procs" is not sorted.
2328
2329- "cgroup.clone_children" is removed.
2330
2331- /proc/cgroups is meaningless for v2.  Use "cgroup.controllers" file
2332  at the root instead.
2333
2334
2335Issues with v1 and Rationales for v2
2336====================================
2337
2338Multiple Hierarchies
2339--------------------
2340
2341cgroup v1 allowed an arbitrary number of hierarchies and each
2342hierarchy could host any number of controllers.  While this seemed to
2343provide a high level of flexibility, it wasn't useful in practice.
2344
2345For example, as there is only one instance of each controller, utility
2346type controllers such as freezer which can be useful in all
2347hierarchies could only be used in one.  The issue is exacerbated by
2348the fact that controllers couldn't be moved to another hierarchy once
2349hierarchies were populated.  Another issue was that all controllers
2350bound to a hierarchy were forced to have exactly the same view of the
2351hierarchy.  It wasn't possible to vary the granularity depending on
2352the specific controller.
2353
2354In practice, these issues heavily limited which controllers could be
2355put on the same hierarchy and most configurations resorted to putting
2356each controller on its own hierarchy.  Only closely related ones, such
2357as the cpu and cpuacct controllers, made sense to be put on the same
2358hierarchy.  This often meant that userland ended up managing multiple
2359similar hierarchies repeating the same steps on each hierarchy
2360whenever a hierarchy management operation was necessary.
2361
2362Furthermore, support for multiple hierarchies came at a steep cost.
2363It greatly complicated cgroup core implementation but more importantly
2364the support for multiple hierarchies restricted how cgroup could be
2365used in general and what controllers was able to do.
2366
2367There was no limit on how many hierarchies there might be, which meant
2368that a thread's cgroup membership couldn't be described in finite
2369length.  The key might contain any number of entries and was unlimited
2370in length, which made it highly awkward to manipulate and led to
2371addition of controllers which existed only to identify membership,
2372which in turn exacerbated the original problem of proliferating number
2373of hierarchies.
2374
2375Also, as a controller couldn't have any expectation regarding the
2376topologies of hierarchies other controllers might be on, each
2377controller had to assume that all other controllers were attached to
2378completely orthogonal hierarchies.  This made it impossible, or at
2379least very cumbersome, for controllers to cooperate with each other.
2380
2381In most use cases, putting controllers on hierarchies which are
2382completely orthogonal to each other isn't necessary.  What usually is
2383called for is the ability to have differing levels of granularity
2384depending on the specific controller.  In other words, hierarchy may
2385be collapsed from leaf towards root when viewed from specific
2386controllers.  For example, a given configuration might not care about
2387how memory is distributed beyond a certain level while still wanting
2388to control how CPU cycles are distributed.
2389
2390
2391Thread Granularity
2392------------------
2393
2394cgroup v1 allowed threads of a process to belong to different cgroups.
2395This didn't make sense for some controllers and those controllers
2396ended up implementing different ways to ignore such situations but
2397much more importantly it blurred the line between API exposed to
2398individual applications and system management interface.
2399
2400Generally, in-process knowledge is available only to the process
2401itself; thus, unlike service-level organization of processes,
2402categorizing threads of a process requires active participation from
2403the application which owns the target process.
2404
2405cgroup v1 had an ambiguously defined delegation model which got abused
2406in combination with thread granularity.  cgroups were delegated to
2407individual applications so that they can create and manage their own
2408sub-hierarchies and control resource distributions along them.  This
2409effectively raised cgroup to the status of a syscall-like API exposed
2410to lay programs.
2411
2412First of all, cgroup has a fundamentally inadequate interface to be
2413exposed this way.  For a process to access its own knobs, it has to
2414extract the path on the target hierarchy from /proc/self/cgroup,
2415construct the path by appending the name of the knob to the path, open
2416and then read and/or write to it.  This is not only extremely clunky
2417and unusual but also inherently racy.  There is no conventional way to
2418define transaction across the required steps and nothing can guarantee
2419that the process would actually be operating on its own sub-hierarchy.
2420
2421cgroup controllers implemented a number of knobs which would never be
2422accepted as public APIs because they were just adding control knobs to
2423system-management pseudo filesystem.  cgroup ended up with interface
2424knobs which were not properly abstracted or refined and directly
2425revealed kernel internal details.  These knobs got exposed to
2426individual applications through the ill-defined delegation mechanism
2427effectively abusing cgroup as a shortcut to implementing public APIs
2428without going through the required scrutiny.
2429
2430This was painful for both userland and kernel.  Userland ended up with
2431misbehaving and poorly abstracted interfaces and kernel exposing and
2432locked into constructs inadvertently.
2433
2434
2435Competition Between Inner Nodes and Threads
2436-------------------------------------------
2437
2438cgroup v1 allowed threads to be in any cgroups which created an
2439interesting problem where threads belonging to a parent cgroup and its
2440children cgroups competed for resources.  This was nasty as two
2441different types of entities competed and there was no obvious way to
2442settle it.  Different controllers did different things.
2443
2444The cpu controller considered threads and cgroups as equivalents and
2445mapped nice levels to cgroup weights.  This worked for some cases but
2446fell flat when children wanted to be allocated specific ratios of CPU
2447cycles and the number of internal threads fluctuated - the ratios
2448constantly changed as the number of competing entities fluctuated.
2449There also were other issues.  The mapping from nice level to weight
2450wasn't obvious or universal, and there were various other knobs which
2451simply weren't available for threads.
2452
2453The io controller implicitly created a hidden leaf node for each
2454cgroup to host the threads.  The hidden leaf had its own copies of all
2455the knobs with ``leaf_`` prefixed.  While this allowed equivalent
2456control over internal threads, it was with serious drawbacks.  It
2457always added an extra layer of nesting which wouldn't be necessary
2458otherwise, made the interface messy and significantly complicated the
2459implementation.
2460
2461The memory controller didn't have a way to control what happened
2462between internal tasks and child cgroups and the behavior was not
2463clearly defined.  There were attempts to add ad-hoc behaviors and
2464knobs to tailor the behavior to specific workloads which would have
2465led to problems extremely difficult to resolve in the long term.
2466
2467Multiple controllers struggled with internal tasks and came up with
2468different ways to deal with it; unfortunately, all the approaches were
2469severely flawed and, furthermore, the widely different behaviors
2470made cgroup as a whole highly inconsistent.
2471
2472This clearly is a problem which needs to be addressed from cgroup core
2473in a uniform way.
2474
2475
2476Other Interface Issues
2477----------------------
2478
2479cgroup v1 grew without oversight and developed a large number of
2480idiosyncrasies and inconsistencies.  One issue on the cgroup core side
2481was how an empty cgroup was notified - a userland helper binary was
2482forked and executed for each event.  The event delivery wasn't
2483recursive or delegatable.  The limitations of the mechanism also led
2484to in-kernel event delivery filtering mechanism further complicating
2485the interface.
2486
2487Controller interfaces were problematic too.  An extreme example is
2488controllers completely ignoring hierarchical organization and treating
2489all cgroups as if they were all located directly under the root
2490cgroup.  Some controllers exposed a large amount of inconsistent
2491implementation details to userland.
2492
2493There also was no consistency across controllers.  When a new cgroup
2494was created, some controllers defaulted to not imposing extra
2495restrictions while others disallowed any resource usage until
2496explicitly configured.  Configuration knobs for the same type of
2497control used widely differing naming schemes and formats.  Statistics
2498and information knobs were named arbitrarily and used different
2499formats and units even in the same controller.
2500
2501cgroup v2 establishes common conventions where appropriate and updates
2502controllers so that they expose minimal and consistent interfaces.
2503
2504
2505Controller Issues and Remedies
2506------------------------------
2507
2508Memory
2509~~~~~~
2510
2511The original lower boundary, the soft limit, is defined as a limit
2512that is per default unset.  As a result, the set of cgroups that
2513global reclaim prefers is opt-in, rather than opt-out.  The costs for
2514optimizing these mostly negative lookups are so high that the
2515implementation, despite its enormous size, does not even provide the
2516basic desirable behavior.  First off, the soft limit has no
2517hierarchical meaning.  All configured groups are organized in a global
2518rbtree and treated like equal peers, regardless where they are located
2519in the hierarchy.  This makes subtree delegation impossible.  Second,
2520the soft limit reclaim pass is so aggressive that it not just
2521introduces high allocation latencies into the system, but also impacts
2522system performance due to overreclaim, to the point where the feature
2523becomes self-defeating.
2524
2525The memory.low boundary on the other hand is a top-down allocated
2526reserve.  A cgroup enjoys reclaim protection when it's within its
2527effective low, which makes delegation of subtrees possible. It also
2528enjoys having reclaim pressure proportional to its overage when
2529above its effective low.
2530
2531The original high boundary, the hard limit, is defined as a strict
2532limit that can not budge, even if the OOM killer has to be called.
2533But this generally goes against the goal of making the most out of the
2534available memory.  The memory consumption of workloads varies during
2535runtime, and that requires users to overcommit.  But doing that with a
2536strict upper limit requires either a fairly accurate prediction of the
2537working set size or adding slack to the limit.  Since working set size
2538estimation is hard and error prone, and getting it wrong results in
2539OOM kills, most users tend to err on the side of a looser limit and
2540end up wasting precious resources.
2541
2542The memory.high boundary on the other hand can be set much more
2543conservatively.  When hit, it throttles allocations by forcing them
2544into direct reclaim to work off the excess, but it never invokes the
2545OOM killer.  As a result, a high boundary that is chosen too
2546aggressively will not terminate the processes, but instead it will
2547lead to gradual performance degradation.  The user can monitor this
2548and make corrections until the minimal memory footprint that still
2549gives acceptable performance is found.
2550
2551In extreme cases, with many concurrent allocations and a complete
2552breakdown of reclaim progress within the group, the high boundary can
2553be exceeded.  But even then it's mostly better to satisfy the
2554allocation from the slack available in other groups or the rest of the
2555system than killing the group.  Otherwise, memory.max is there to
2556limit this type of spillover and ultimately contain buggy or even
2557malicious applications.
2558
2559Setting the original memory.limit_in_bytes below the current usage was
2560subject to a race condition, where concurrent charges could cause the
2561limit setting to fail. memory.max on the other hand will first set the
2562limit to prevent new charges, and then reclaim and OOM kill until the
2563new limit is met - or the task writing to memory.max is killed.
2564
2565The combined memory+swap accounting and limiting is replaced by real
2566control over swap space.
2567
2568The main argument for a combined memory+swap facility in the original
2569cgroup design was that global or parental pressure would always be
2570able to swap all anonymous memory of a child group, regardless of the
2571child's own (possibly untrusted) configuration.  However, untrusted
2572groups can sabotage swapping by other means - such as referencing its
2573anonymous memory in a tight loop - and an admin can not assume full
2574swappability when overcommitting untrusted jobs.
2575
2576For trusted jobs, on the other hand, a combined counter is not an
2577intuitive userspace interface, and it flies in the face of the idea
2578that cgroup controllers should account and limit specific physical
2579resources.  Swap space is a resource like all others in the system,
2580and that's why unified hierarchy allows distributing it separately.
2581