xref: /linux/Documentation/admin-guide/cgroup-v2.rst (revision 7d1f84f1b94e79cb95d974ab3ec54cb3a4d64f5b)
1.. _cgroup-v2:
2
3================
4Control Group v2
5================
6
7:Date: October, 2015
8:Author: Tejun Heo <tj@kernel.org>
9
10This is the authoritative documentation on the design, interface and
11conventions of cgroup v2.  It describes all userland-visible aspects
12of cgroup including core and specific controller behaviors.  All
13future changes must be reflected in this document.  Documentation for
14v1 is available under :ref:`Documentation/admin-guide/cgroup-v1/index.rst <cgroup-v1>`.
15
16.. CONTENTS
17
18   [Whenever any new section is added to this document, please also add
19    an entry here.]
20
21   1. Introduction
22     1-1. Terminology
23     1-2. What is cgroup?
24   2. Basic Operations
25     2-1. Mounting
26     2-2. Organizing Processes and Threads
27       2-2-1. Processes
28       2-2-2. Threads
29     2-3. [Un]populated Notification
30     2-4. Controlling Controllers
31       2-4-1. Availability
32       2-4-2. Enabling and Disabling
33       2-4-3. Top-down Constraint
34       2-4-4. No Internal Process Constraint
35     2-5. Delegation
36       2-5-1. Model of Delegation
37       2-5-2. Delegation Containment
38     2-6. Guidelines
39       2-6-1. Organize Once and Control
40       2-6-2. Avoid Name Collisions
41   3. Resource Distribution Models
42     3-1. Weights
43     3-2. Limits
44     3-3. Protections
45     3-4. Allocations
46   4. Interface Files
47     4-1. Format
48     4-2. Conventions
49     4-3. Core Interface Files
50   5. Controllers
51     5-1. CPU
52       5-1-1. CPU Interface Files
53     5-2. Memory
54       5-2-1. Memory Interface Files
55       5-2-2. Usage Guidelines
56       5-2-3. Reclaim Protection
57       5-2-4. Memory Ownership
58     5-3. IO
59       5-3-1. IO Interface Files
60       5-3-2. Writeback
61       5-3-3. IO Latency
62         5-3-3-1. How IO Latency Throttling Works
63         5-3-3-2. IO Latency Interface Files
64       5-3-4. IO Priority
65     5-4. PID
66       5-4-1. PID Interface Files
67     5-5. Cpuset
68       5.5-1. Cpuset Interface Files
69     5-6. Device controller
70     5-7. RDMA
71       5-7-1. RDMA Interface Files
72     5-8. DMEM
73       5-8-1. DMEM Interface Files
74     5-9. HugeTLB
75       5.9-1. HugeTLB Interface Files
76     5-10. Misc
77       5.10-1 Misc Interface Files
78       5.10-2 Migration and Ownership
79     5-11. Others
80       5-11-1. perf_event
81     5-N. Non-normative information
82       5-N-1. CPU controller root cgroup process behaviour
83       5-N-2. IO controller root cgroup process behaviour
84   6. Namespace
85     6-1. Basics
86     6-2. The Root and Views
87     6-3. Migration and setns(2)
88     6-4. Interaction with Other Namespaces
89   P. Information on Kernel Programming
90     P-1. Filesystem Support for Writeback
91   D. Deprecated v1 Core Features
92   R. Issues with v1 and Rationales for v2
93     R-1. Multiple Hierarchies
94     R-2. Thread Granularity
95     R-3. Competition Between Inner Nodes and Threads
96     R-4. Other Interface Issues
97     R-5. Controller Issues and Remedies
98       R-5-1. Memory
99
100
101Introduction
102============
103
104Terminology
105-----------
106
107"cgroup" stands for "control group" and is never capitalized.  The
108singular form is used to designate the whole feature and also as a
109qualifier as in "cgroup controllers".  When explicitly referring to
110multiple individual control groups, the plural form "cgroups" is used.
111
112
113What is cgroup?
114---------------
115
116cgroup is a mechanism to organize processes hierarchically and
117distribute system resources along the hierarchy in a controlled and
118configurable manner.
119
120cgroup is largely composed of two parts - the core and controllers.
121cgroup core is primarily responsible for hierarchically organizing
122processes.  A cgroup controller is usually responsible for
123distributing a specific type of system resource along the hierarchy
124although there are utility controllers which serve purposes other than
125resource distribution.
126
127cgroups form a tree structure and every process in the system belongs
128to one and only one cgroup.  All threads of a process belong to the
129same cgroup.  On creation, all processes are put in the cgroup that
130the parent process belongs to at the time.  A process can be migrated
131to another cgroup.  Migration of a process doesn't affect already
132existing descendant processes.
133
134Following certain structural constraints, controllers may be enabled or
135disabled selectively on a cgroup.  All controller behaviors are
136hierarchical - if a controller is enabled on a cgroup, it affects all
137processes which belong to the cgroups consisting the inclusive
138sub-hierarchy of the cgroup.  When a controller is enabled on a nested
139cgroup, it always restricts the resource distribution further.  The
140restrictions set closer to the root in the hierarchy can not be
141overridden from further away.
142
143
144Basic Operations
145================
146
147Mounting
148--------
149
150Unlike v1, cgroup v2 has only single hierarchy.  The cgroup v2
151hierarchy can be mounted with the following mount command::
152
153  # mount -t cgroup2 none $MOUNT_POINT
154
155cgroup2 filesystem has the magic number 0x63677270 ("cgrp").  All
156controllers which support v2 and are not bound to a v1 hierarchy are
157automatically bound to the v2 hierarchy and show up at the root.
158Controllers which are not in active use in the v2 hierarchy can be
159bound to other hierarchies.  This allows mixing v2 hierarchy with the
160legacy v1 multiple hierarchies in a fully backward compatible way.
161
162A controller can be moved across hierarchies only after the controller
163is no longer referenced in its current hierarchy.  Because per-cgroup
164controller states are destroyed asynchronously and controllers may
165have lingering references, a controller may not show up immediately on
166the v2 hierarchy after the final umount of the previous hierarchy.
167Similarly, a controller should be fully disabled to be moved out of
168the unified hierarchy and it may take some time for the disabled
169controller to become available for other hierarchies; furthermore, due
170to inter-controller dependencies, other controllers may need to be
171disabled too.
172
173While useful for development and manual configurations, moving
174controllers dynamically between the v2 and other hierarchies is
175strongly discouraged for production use.  It is recommended to decide
176the hierarchies and controller associations before starting using the
177controllers after system boot.
178
179During transition to v2, system management software might still
180automount the v1 cgroup filesystem and so hijack all controllers
181during boot, before manual intervention is possible. To make testing
182and experimenting easier, the kernel parameter cgroup_no_v1= allows
183disabling controllers in v1 and make them always available in v2.
184
185cgroup v2 currently supports the following mount options.
186
187  nsdelegate
188	Consider cgroup namespaces as delegation boundaries.  This
189	option is system wide and can only be set on mount or modified
190	through remount from the init namespace.  The mount option is
191	ignored on non-init namespace mounts.  Please refer to the
192	Delegation section for details.
193
194  favordynmods
195        Reduce the latencies of dynamic cgroup modifications such as
196        task migrations and controller on/offs at the cost of making
197        hot path operations such as forks and exits more expensive.
198        The static usage pattern of creating a cgroup, enabling
199        controllers, and then seeding it with CLONE_INTO_CGROUP is
200        not affected by this option.
201
202  memory_localevents
203        Only populate memory.events with data for the current cgroup,
204        and not any subtrees. This is legacy behaviour, the default
205        behaviour without this option is to include subtree counts.
206        This option is system wide and can only be set on mount or
207        modified through remount from the init namespace. The mount
208        option is ignored on non-init namespace mounts.
209
210  memory_recursiveprot
211        Recursively apply memory.min and memory.low protection to
212        entire subtrees, without requiring explicit downward
213        propagation into leaf cgroups.  This allows protecting entire
214        subtrees from one another, while retaining free competition
215        within those subtrees.  This should have been the default
216        behavior but is a mount-option to avoid regressing setups
217        relying on the original semantics (e.g. specifying bogusly
218        high 'bypass' protection values at higher tree levels).
219
220  memory_hugetlb_accounting
221        Count HugeTLB memory usage towards the cgroup's overall
222        memory usage for the memory controller (for the purpose of
223        statistics reporting and memory protection). This is a new
224        behavior that could regress existing setups, so it must be
225        explicitly opted in with this mount option.
226
227        A few caveats to keep in mind:
228
229        * There is no HugeTLB pool management involved in the memory
230          controller. The pre-allocated pool does not belong to anyone.
231          Specifically, when a new HugeTLB folio is allocated to
232          the pool, it is not accounted for from the perspective of the
233          memory controller. It is only charged to a cgroup when it is
234          actually used (for e.g at page fault time). Host memory
235          overcommit management has to consider this when configuring
236          hard limits. In general, HugeTLB pool management should be
237          done via other mechanisms (such as the HugeTLB controller).
238        * Failure to charge a HugeTLB folio to the memory controller
239          results in SIGBUS. This could happen even if the HugeTLB pool
240          still has pages available (but the cgroup limit is hit and
241          reclaim attempt fails).
242        * Charging HugeTLB memory towards the memory controller affects
243          memory protection and reclaim dynamics. Any userspace tuning
244          (of low, min limits for e.g) needs to take this into account.
245        * HugeTLB pages utilized while this option is not selected
246          will not be tracked by the memory controller (even if cgroup
247          v2 is remounted later on).
248
249  pids_localevents
250        The option restores v1-like behavior of pids.events:max, that is only
251        local (inside cgroup proper) fork failures are counted. Without this
252        option pids.events.max represents any pids.max enforcemnt across
253        cgroup's subtree.
254
255
256
257Organizing Processes and Threads
258--------------------------------
259
260Processes
261~~~~~~~~~
262
263Initially, only the root cgroup exists to which all processes belong.
264A child cgroup can be created by creating a sub-directory::
265
266  # mkdir $CGROUP_NAME
267
268A given cgroup may have multiple child cgroups forming a tree
269structure.  Each cgroup has a read-writable interface file
270"cgroup.procs".  When read, it lists the PIDs of all processes which
271belong to the cgroup one-per-line.  The PIDs are not ordered and the
272same PID may show up more than once if the process got moved to
273another cgroup and then back or the PID got recycled while reading.
274
275A process can be migrated into a cgroup by writing its PID to the
276target cgroup's "cgroup.procs" file.  Only one process can be migrated
277on a single write(2) call.  If a process is composed of multiple
278threads, writing the PID of any thread migrates all threads of the
279process.
280
281When a process forks a child process, the new process is born into the
282cgroup that the forking process belongs to at the time of the
283operation.  After exit, a process stays associated with the cgroup
284that it belonged to at the time of exit until it's reaped; however, a
285zombie process does not appear in "cgroup.procs" and thus can't be
286moved to another cgroup.
287
288A cgroup which doesn't have any children or live processes can be
289destroyed by removing the directory.  Note that a cgroup which doesn't
290have any children and is associated only with zombie processes is
291considered empty and can be removed::
292
293  # rmdir $CGROUP_NAME
294
295"/proc/$PID/cgroup" lists a process's cgroup membership.  If legacy
296cgroup is in use in the system, this file may contain multiple lines,
297one for each hierarchy.  The entry for cgroup v2 is always in the
298format "0::$PATH"::
299
300  # cat /proc/842/cgroup
301  ...
302  0::/test-cgroup/test-cgroup-nested
303
304If the process becomes a zombie and the cgroup it was associated with
305is removed subsequently, " (deleted)" is appended to the path::
306
307  # cat /proc/842/cgroup
308  ...
309  0::/test-cgroup/test-cgroup-nested (deleted)
310
311
312Threads
313~~~~~~~
314
315cgroup v2 supports thread granularity for a subset of controllers to
316support use cases requiring hierarchical resource distribution across
317the threads of a group of processes.  By default, all threads of a
318process belong to the same cgroup, which also serves as the resource
319domain to host resource consumptions which are not specific to a
320process or thread.  The thread mode allows threads to be spread across
321a subtree while still maintaining the common resource domain for them.
322
323Controllers which support thread mode are called threaded controllers.
324The ones which don't are called domain controllers.
325
326Marking a cgroup threaded makes it join the resource domain of its
327parent as a threaded cgroup.  The parent may be another threaded
328cgroup whose resource domain is further up in the hierarchy.  The root
329of a threaded subtree, that is, the nearest ancestor which is not
330threaded, is called threaded domain or thread root interchangeably and
331serves as the resource domain for the entire subtree.
332
333Inside a threaded subtree, threads of a process can be put in
334different cgroups and are not subject to the no internal process
335constraint - threaded controllers can be enabled on non-leaf cgroups
336whether they have threads in them or not.
337
338As the threaded domain cgroup hosts all the domain resource
339consumptions of the subtree, it is considered to have internal
340resource consumptions whether there are processes in it or not and
341can't have populated child cgroups which aren't threaded.  Because the
342root cgroup is not subject to no internal process constraint, it can
343serve both as a threaded domain and a parent to domain cgroups.
344
345The current operation mode or type of the cgroup is shown in the
346"cgroup.type" file which indicates whether the cgroup is a normal
347domain, a domain which is serving as the domain of a threaded subtree,
348or a threaded cgroup.
349
350On creation, a cgroup is always a domain cgroup and can be made
351threaded by writing "threaded" to the "cgroup.type" file.  The
352operation is single direction::
353
354  # echo threaded > cgroup.type
355
356Once threaded, the cgroup can't be made a domain again.  To enable the
357thread mode, the following conditions must be met.
358
359- As the cgroup will join the parent's resource domain.  The parent
360  must either be a valid (threaded) domain or a threaded cgroup.
361
362- When the parent is an unthreaded domain, it must not have any domain
363  controllers enabled or populated domain children.  The root is
364  exempt from this requirement.
365
366Topology-wise, a cgroup can be in an invalid state.  Please consider
367the following topology::
368
369  A (threaded domain) - B (threaded) - C (domain, just created)
370
371C is created as a domain but isn't connected to a parent which can
372host child domains.  C can't be used until it is turned into a
373threaded cgroup.  "cgroup.type" file will report "domain (invalid)" in
374these cases.  Operations which fail due to invalid topology use
375EOPNOTSUPP as the errno.
376
377A domain cgroup is turned into a threaded domain when one of its child
378cgroup becomes threaded or threaded controllers are enabled in the
379"cgroup.subtree_control" file while there are processes in the cgroup.
380A threaded domain reverts to a normal domain when the conditions
381clear.
382
383When read, "cgroup.threads" contains the list of the thread IDs of all
384threads in the cgroup.  Except that the operations are per-thread
385instead of per-process, "cgroup.threads" has the same format and
386behaves the same way as "cgroup.procs".  While "cgroup.threads" can be
387written to in any cgroup, as it can only move threads inside the same
388threaded domain, its operations are confined inside each threaded
389subtree.
390
391The threaded domain cgroup serves as the resource domain for the whole
392subtree, and, while the threads can be scattered across the subtree,
393all the processes are considered to be in the threaded domain cgroup.
394"cgroup.procs" in a threaded domain cgroup contains the PIDs of all
395processes in the subtree and is not readable in the subtree proper.
396However, "cgroup.procs" can be written to from anywhere in the subtree
397to migrate all threads of the matching process to the cgroup.
398
399Only threaded controllers can be enabled in a threaded subtree.  When
400a threaded controller is enabled inside a threaded subtree, it only
401accounts for and controls resource consumptions associated with the
402threads in the cgroup and its descendants.  All consumptions which
403aren't tied to a specific thread belong to the threaded domain cgroup.
404
405Because a threaded subtree is exempt from no internal process
406constraint, a threaded controller must be able to handle competition
407between threads in a non-leaf cgroup and its child cgroups.  Each
408threaded controller defines how such competitions are handled.
409
410Currently, the following controllers are threaded and can be enabled
411in a threaded cgroup::
412
413- cpu
414- cpuset
415- perf_event
416- pids
417
418[Un]populated Notification
419--------------------------
420
421Each non-root cgroup has a "cgroup.events" file which contains
422"populated" field indicating whether the cgroup's sub-hierarchy has
423live processes in it.  Its value is 0 if there is no live process in
424the cgroup and its descendants; otherwise, 1.  poll and [id]notify
425events are triggered when the value changes.  This can be used, for
426example, to start a clean-up operation after all processes of a given
427sub-hierarchy have exited.  The populated state updates and
428notifications are recursive.  Consider the following sub-hierarchy
429where the numbers in the parentheses represent the numbers of processes
430in each cgroup::
431
432  A(4) - B(0) - C(1)
433              \ D(0)
434
435A, B and C's "populated" fields would be 1 while D's 0.  After the one
436process in C exits, B and C's "populated" fields would flip to "0" and
437file modified events will be generated on the "cgroup.events" files of
438both cgroups.
439
440
441Controlling Controllers
442-----------------------
443
444Availability
445~~~~~~~~~~~~
446
447A controller is available in a cgroup when it is supported by the kernel (i.e.,
448compiled in, not disabled and not attached to a v1 hierarchy) and listed in the
449"cgroup.controllers" file. Availability means the controller's interface files
450are exposed in the cgroup’s directory, allowing the distribution of the target
451resource to be observed or controlled within that cgroup.
452
453Enabling and Disabling
454~~~~~~~~~~~~~~~~~~~~~~
455
456Each cgroup has a "cgroup.controllers" file which lists all
457controllers available for the cgroup to enable::
458
459  # cat cgroup.controllers
460  cpu io memory
461
462No controller is enabled by default.  Controllers can be enabled and
463disabled by writing to the "cgroup.subtree_control" file::
464
465  # echo "+cpu +memory -io" > cgroup.subtree_control
466
467Only controllers which are listed in "cgroup.controllers" can be
468enabled.  When multiple operations are specified as above, either they
469all succeed or fail.  If multiple operations on the same controller
470are specified, the last one is effective.
471
472Enabling a controller in a cgroup indicates that the distribution of
473the target resource across its immediate children will be controlled.
474Consider the following sub-hierarchy.  The enabled controllers are
475listed in parentheses::
476
477  A(cpu,memory) - B(memory) - C()
478                            \ D()
479
480As A has "cpu" and "memory" enabled, A will control the distribution
481of CPU cycles and memory to its children, in this case, B.  As B has
482"memory" enabled but not "CPU", C and D will compete freely on CPU
483cycles but their division of memory available to B will be controlled.
484
485As a controller regulates the distribution of the target resource to
486the cgroup's children, enabling it creates the controller's interface
487files in the child cgroups.  In the above example, enabling "cpu" on B
488would create the "cpu." prefixed controller interface files in C and
489D.  Likewise, disabling "memory" from B would remove the "memory."
490prefixed controller interface files from C and D.  This means that the
491controller interface files - anything which doesn't start with
492"cgroup." are owned by the parent rather than the cgroup itself.
493
494
495Top-down Constraint
496~~~~~~~~~~~~~~~~~~~
497
498Resources are distributed top-down and a cgroup can further distribute
499a resource only if the resource has been distributed to it from the
500parent.  This means that all non-root "cgroup.subtree_control" files
501can only contain controllers which are enabled in the parent's
502"cgroup.subtree_control" file.  A controller can be enabled only if
503the parent has the controller enabled and a controller can't be
504disabled if one or more children have it enabled.
505
506
507No Internal Process Constraint
508~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
509
510Non-root cgroups can distribute domain resources to their children
511only when they don't have any processes of their own.  In other words,
512only domain cgroups which don't contain any processes can have domain
513controllers enabled in their "cgroup.subtree_control" files.
514
515This guarantees that, when a domain controller is looking at the part
516of the hierarchy which has it enabled, processes are always only on
517the leaves.  This rules out situations where child cgroups compete
518against internal processes of the parent.
519
520The root cgroup is exempt from this restriction.  Root contains
521processes and anonymous resource consumption which can't be associated
522with any other cgroups and requires special treatment from most
523controllers.  How resource consumption in the root cgroup is governed
524is up to each controller (for more information on this topic please
525refer to the Non-normative information section in the Controllers
526chapter).
527
528Note that the restriction doesn't get in the way if there is no
529enabled controller in the cgroup's "cgroup.subtree_control".  This is
530important as otherwise it wouldn't be possible to create children of a
531populated cgroup.  To control resource distribution of a cgroup, the
532cgroup must create children and transfer all its processes to the
533children before enabling controllers in its "cgroup.subtree_control"
534file.
535
536
537Delegation
538----------
539
540Model of Delegation
541~~~~~~~~~~~~~~~~~~~
542
543A cgroup can be delegated in two ways.  First, to a less privileged
544user by granting write access of the directory and its "cgroup.procs",
545"cgroup.threads" and "cgroup.subtree_control" files to the user.
546Second, if the "nsdelegate" mount option is set, automatically to a
547cgroup namespace on namespace creation.
548
549Because the resource control interface files in a given directory
550control the distribution of the parent's resources, the delegatee
551shouldn't be allowed to write to them.  For the first method, this is
552achieved by not granting access to these files.  For the second, files
553outside the namespace should be hidden from the delegatee by the means
554of at least mount namespacing, and the kernel rejects writes to all
555files on a namespace root from inside the cgroup namespace, except for
556those files listed in "/sys/kernel/cgroup/delegate" (including
557"cgroup.procs", "cgroup.threads", "cgroup.subtree_control", etc.).
558
559The end results are equivalent for both delegation types.  Once
560delegated, the user can build sub-hierarchy under the directory,
561organize processes inside it as it sees fit and further distribute the
562resources it received from the parent.  The limits and other settings
563of all resource controllers are hierarchical and regardless of what
564happens in the delegated sub-hierarchy, nothing can escape the
565resource restrictions imposed by the parent.
566
567Currently, cgroup doesn't impose any restrictions on the number of
568cgroups in or nesting depth of a delegated sub-hierarchy; however,
569this may be limited explicitly in the future.
570
571
572Delegation Containment
573~~~~~~~~~~~~~~~~~~~~~~
574
575A delegated sub-hierarchy is contained in the sense that processes
576can't be moved into or out of the sub-hierarchy by the delegatee.
577
578For delegations to a less privileged user, this is achieved by
579requiring the following conditions for a process with a non-root euid
580to migrate a target process into a cgroup by writing its PID to the
581"cgroup.procs" file.
582
583- The writer must have write access to the "cgroup.procs" file.
584
585- The writer must have write access to the "cgroup.procs" file of the
586  common ancestor of the source and destination cgroups.
587
588The above two constraints ensure that while a delegatee may migrate
589processes around freely in the delegated sub-hierarchy it can't pull
590in from or push out to outside the sub-hierarchy.
591
592For an example, let's assume cgroups C0 and C1 have been delegated to
593user U0 who created C00, C01 under C0 and C10 under C1 as follows and
594all processes under C0 and C1 belong to U0::
595
596  ~~~~~~~~~~~~~ - C0 - C00
597  ~ cgroup    ~      \ C01
598  ~ hierarchy ~
599  ~~~~~~~~~~~~~ - C1 - C10
600
601Let's also say U0 wants to write the PID of a process which is
602currently in C10 into "C00/cgroup.procs".  U0 has write access to the
603file; however, the common ancestor of the source cgroup C10 and the
604destination cgroup C00 is above the points of delegation and U0 would
605not have write access to its "cgroup.procs" files and thus the write
606will be denied with -EACCES.
607
608For delegations to namespaces, containment is achieved by requiring
609that both the source and destination cgroups are reachable from the
610namespace of the process which is attempting the migration.  If either
611is not reachable, the migration is rejected with -ENOENT.
612
613
614Guidelines
615----------
616
617Organize Once and Control
618~~~~~~~~~~~~~~~~~~~~~~~~~
619
620Migrating a process across cgroups is a relatively expensive operation
621and stateful resources such as memory are not moved together with the
622process.  This is an explicit design decision as there often exist
623inherent trade-offs between migration and various hot paths in terms
624of synchronization cost.
625
626As such, migrating processes across cgroups frequently as a means to
627apply different resource restrictions is discouraged.  A workload
628should be assigned to a cgroup according to the system's logical and
629resource structure once on start-up.  Dynamic adjustments to resource
630distribution can be made by changing controller configuration through
631the interface files.
632
633
634Avoid Name Collisions
635~~~~~~~~~~~~~~~~~~~~~
636
637Interface files for a cgroup and its children cgroups occupy the same
638directory and it is possible to create children cgroups which collide
639with interface files.
640
641All cgroup core interface files are prefixed with "cgroup." and each
642controller's interface files are prefixed with the controller name and
643a dot.  A controller's name is composed of lower case alphabets and
644'_'s but never begins with an '_' so it can be used as the prefix
645character for collision avoidance.  Also, interface file names won't
646start or end with terms which are often used in categorizing workloads
647such as job, service, slice, unit or workload.
648
649cgroup doesn't do anything to prevent name collisions and it's the
650user's responsibility to avoid them.
651
652
653Resource Distribution Models
654============================
655
656cgroup controllers implement several resource distribution schemes
657depending on the resource type and expected use cases.  This section
658describes major schemes in use along with their expected behaviors.
659
660
661Weights
662-------
663
664A parent's resource is distributed by adding up the weights of all
665active children and giving each the fraction matching the ratio of its
666weight against the sum.  As only children which can make use of the
667resource at the moment participate in the distribution, this is
668work-conserving.  Due to the dynamic nature, this model is usually
669used for stateless resources.
670
671All weights are in the range [1, 10000] with the default at 100.  This
672allows symmetric multiplicative biases in both directions at fine
673enough granularity while staying in the intuitive range.
674
675As long as the weight is in range, all configuration combinations are
676valid and there is no reason to reject configuration changes or
677process migrations.
678
679"cpu.weight" proportionally distributes CPU cycles to active children
680and is an example of this type.
681
682
683.. _cgroupv2-limits-distributor:
684
685Limits
686------
687
688A child can only consume up to the configured amount of the resource.
689Limits can be over-committed - the sum of the limits of children can
690exceed the amount of resource available to the parent.
691
692Limits are in the range [0, max] and defaults to "max", which is noop.
693
694As limits can be over-committed, all configuration combinations are
695valid and there is no reason to reject configuration changes or
696process migrations.
697
698"io.max" limits the maximum BPS and/or IOPS that a cgroup can consume
699on an IO device and is an example of this type.
700
701.. _cgroupv2-protections-distributor:
702
703Protections
704-----------
705
706A cgroup is protected up to the configured amount of the resource
707as long as the usages of all its ancestors are under their
708protected levels.  Protections can be hard guarantees or best effort
709soft boundaries.  Protections can also be over-committed in which case
710only up to the amount available to the parent is protected among
711children.
712
713Protections are in the range [0, max] and defaults to 0, which is
714noop.
715
716As protections can be over-committed, all configuration combinations
717are valid and there is no reason to reject configuration changes or
718process migrations.
719
720"memory.low" implements best-effort memory protection and is an
721example of this type.
722
723
724Allocations
725-----------
726
727A cgroup is exclusively allocated a certain amount of a finite
728resource.  Allocations can't be over-committed - the sum of the
729allocations of children can not exceed the amount of resource
730available to the parent.
731
732Allocations are in the range [0, max] and defaults to 0, which is no
733resource.
734
735As allocations can't be over-committed, some configuration
736combinations are invalid and should be rejected.  Also, if the
737resource is mandatory for execution of processes, process migrations
738may be rejected.
739
740
741Interface Files
742===============
743
744Format
745------
746
747All interface files should be in one of the following formats whenever
748possible::
749
750  New-line separated values
751  (when only one value can be written at once)
752
753	VAL0\n
754	VAL1\n
755	...
756
757  Space separated values
758  (when read-only or multiple values can be written at once)
759
760	VAL0 VAL1 ...\n
761
762  Flat keyed
763
764	KEY0 VAL0\n
765	KEY1 VAL1\n
766	...
767
768  Nested keyed
769
770	KEY0 SUB_KEY0=VAL00 SUB_KEY1=VAL01...
771	KEY1 SUB_KEY0=VAL10 SUB_KEY1=VAL11...
772	...
773
774For a writable file, the format for writing should generally match
775reading; however, controllers may allow omitting later fields or
776implement restricted shortcuts for most common use cases.
777
778For both flat and nested keyed files, only the values for a single key
779can be written at a time.  For nested keyed files, the sub key pairs
780may be specified in any order and not all pairs have to be specified.
781
782
783Conventions
784-----------
785
786- Settings for a single feature should be contained in a single file.
787
788- The root cgroup should be exempt from resource control and thus
789  shouldn't have resource control interface files.
790
791- The default time unit is microseconds.  If a different unit is ever
792  used, an explicit unit suffix must be present.
793
794- A parts-per quantity should use a percentage decimal with at least
795  two digit fractional part - e.g. 13.40.
796
797- If a controller implements weight based resource distribution, its
798  interface file should be named "weight" and have the range [1,
799  10000] with 100 as the default.  The values are chosen to allow
800  enough and symmetric bias in both directions while keeping it
801  intuitive (the default is 100%).
802
803- If a controller implements an absolute resource guarantee and/or
804  limit, the interface files should be named "min" and "max"
805  respectively.  If a controller implements best effort resource
806  guarantee and/or limit, the interface files should be named "low"
807  and "high" respectively.
808
809  In the above four control files, the special token "max" should be
810  used to represent upward infinity for both reading and writing.
811
812- If a setting has a configurable default value and keyed specific
813  overrides, the default entry should be keyed with "default" and
814  appear as the first entry in the file.
815
816  The default value can be updated by writing either "default $VAL" or
817  "$VAL".
818
819  When writing to update a specific override, "default" can be used as
820  the value to indicate removal of the override.  Override entries
821  with "default" as the value must not appear when read.
822
823  For example, a setting which is keyed by major:minor device numbers
824  with integer values may look like the following::
825
826    # cat cgroup-example-interface-file
827    default 150
828    8:0 300
829
830  The default value can be updated by::
831
832    # echo 125 > cgroup-example-interface-file
833
834  or::
835
836    # echo "default 125" > cgroup-example-interface-file
837
838  An override can be set by::
839
840    # echo "8:16 170" > cgroup-example-interface-file
841
842  and cleared by::
843
844    # echo "8:0 default" > cgroup-example-interface-file
845    # cat cgroup-example-interface-file
846    default 125
847    8:16 170
848
849- For events which are not very high frequency, an interface file
850  "events" should be created which lists event key value pairs.
851  Whenever a notifiable event happens, file modified event should be
852  generated on the file.
853
854
855Core Interface Files
856--------------------
857
858All cgroup core files are prefixed with "cgroup."
859
860  cgroup.type
861	A read-write single value file which exists on non-root
862	cgroups.
863
864	When read, it indicates the current type of the cgroup, which
865	can be one of the following values.
866
867	- "domain" : A normal valid domain cgroup.
868
869	- "domain threaded" : A threaded domain cgroup which is
870          serving as the root of a threaded subtree.
871
872	- "domain invalid" : A cgroup which is in an invalid state.
873	  It can't be populated or have controllers enabled.  It may
874	  be allowed to become a threaded cgroup.
875
876	- "threaded" : A threaded cgroup which is a member of a
877          threaded subtree.
878
879	A cgroup can be turned into a threaded cgroup by writing
880	"threaded" to this file.
881
882  cgroup.procs
883	A read-write new-line separated values file which exists on
884	all cgroups.
885
886	When read, it lists the PIDs of all processes which belong to
887	the cgroup one-per-line.  The PIDs are not ordered and the
888	same PID may show up more than once if the process got moved
889	to another cgroup and then back or the PID got recycled while
890	reading.
891
892	A PID can be written to migrate the process associated with
893	the PID to the cgroup.  The writer should match all of the
894	following conditions.
895
896	- It must have write access to the "cgroup.procs" file.
897
898	- It must have write access to the "cgroup.procs" file of the
899	  common ancestor of the source and destination cgroups.
900
901	When delegating a sub-hierarchy, write access to this file
902	should be granted along with the containing directory.
903
904	In a threaded cgroup, reading this file fails with EOPNOTSUPP
905	as all the processes belong to the thread root.  Writing is
906	supported and moves every thread of the process to the cgroup.
907
908  cgroup.threads
909	A read-write new-line separated values file which exists on
910	all cgroups.
911
912	When read, it lists the TIDs of all threads which belong to
913	the cgroup one-per-line.  The TIDs are not ordered and the
914	same TID may show up more than once if the thread got moved to
915	another cgroup and then back or the TID got recycled while
916	reading.
917
918	A TID can be written to migrate the thread associated with the
919	TID to the cgroup.  The writer should match all of the
920	following conditions.
921
922	- It must have write access to the "cgroup.threads" file.
923
924	- The cgroup that the thread is currently in must be in the
925          same resource domain as the destination cgroup.
926
927	- It must have write access to the "cgroup.procs" file of the
928	  common ancestor of the source and destination cgroups.
929
930	When delegating a sub-hierarchy, write access to this file
931	should be granted along with the containing directory.
932
933  cgroup.controllers
934	A read-only space separated values file which exists on all
935	cgroups.
936
937	It shows space separated list of all controllers available to
938	the cgroup.  The controllers are not ordered.
939
940  cgroup.subtree_control
941	A read-write space separated values file which exists on all
942	cgroups.  Starts out empty.
943
944	When read, it shows space separated list of the controllers
945	which are enabled to control resource distribution from the
946	cgroup to its children.
947
948	Space separated list of controllers prefixed with '+' or '-'
949	can be written to enable or disable controllers.  A controller
950	name prefixed with '+' enables the controller and '-'
951	disables.  If a controller appears more than once on the list,
952	the last one is effective.  When multiple enable and disable
953	operations are specified, either all succeed or all fail.
954
955  cgroup.events
956	A read-only flat-keyed file which exists on non-root cgroups.
957	The following entries are defined.  Unless specified
958	otherwise, a value change in this file generates a file
959	modified event.
960
961	  populated
962		1 if the cgroup or its descendants contains any live
963		processes; otherwise, 0.
964	  frozen
965		1 if the cgroup is frozen; otherwise, 0.
966
967  cgroup.max.descendants
968	A read-write single value files.  The default is "max".
969
970	Maximum allowed number of descent cgroups.
971	If the actual number of descendants is equal or larger,
972	an attempt to create a new cgroup in the hierarchy will fail.
973
974  cgroup.max.depth
975	A read-write single value files.  The default is "max".
976
977	Maximum allowed descent depth below the current cgroup.
978	If the actual descent depth is equal or larger,
979	an attempt to create a new child cgroup will fail.
980
981  cgroup.stat
982	A read-only flat-keyed file with the following entries:
983
984	  nr_descendants
985		Total number of visible descendant cgroups.
986
987	  nr_dying_descendants
988		Total number of dying descendant cgroups. A cgroup becomes
989		dying after being deleted by a user. The cgroup will remain
990		in dying state for some time undefined time (which can depend
991		on system load) before being completely destroyed.
992
993		A process can't enter a dying cgroup under any circumstances,
994		a dying cgroup can't revive.
995
996		A dying cgroup can consume system resources not exceeding
997		limits, which were active at the moment of cgroup deletion.
998
999	  nr_subsys_<cgroup_subsys>
1000		Total number of live cgroup subsystems (e.g memory
1001		cgroup) at and beneath the current cgroup.
1002
1003	  nr_dying_subsys_<cgroup_subsys>
1004		Total number of dying cgroup subsystems (e.g. memory
1005		cgroup) at and beneath the current cgroup.
1006
1007  cgroup.stat.local
1008	A read-only flat-keyed file which exists in non-root cgroups.
1009	The following entry is defined:
1010
1011	  frozen_usec
1012		Cumulative time that this cgroup has spent between freezing and
1013		thawing, regardless of whether by self or ancestor groups.
1014		NB: (not) reaching "frozen" state is not accounted here.
1015
1016		Using the following ASCII representation of a cgroup's freezer
1017		state, ::
1018
1019			       1    _____
1020			frozen 0 __/     \__
1021			          ab    cd
1022
1023		the duration being measured is the span between a and c.
1024
1025  cgroup.freeze
1026	A read-write single value file which exists on non-root cgroups.
1027	Allowed values are "0" and "1". The default is "0".
1028
1029	Writing "1" to the file causes freezing of the cgroup and all
1030	descendant cgroups. This means that all belonging processes will
1031	be stopped and will not run until the cgroup will be explicitly
1032	unfrozen. Freezing of the cgroup may take some time; when this action
1033	is completed, the "frozen" value in the cgroup.events control file
1034	will be updated to "1" and the corresponding notification will be
1035	issued.
1036
1037	A cgroup can be frozen either by its own settings, or by settings
1038	of any ancestor cgroups. If any of ancestor cgroups is frozen, the
1039	cgroup will remain frozen.
1040
1041	Processes in the frozen cgroup can be killed by a fatal signal.
1042	They also can enter and leave a frozen cgroup: either by an explicit
1043	move by a user, or if freezing of the cgroup races with fork().
1044	If a process is moved to a frozen cgroup, it stops. If a process is
1045	moved out of a frozen cgroup, it becomes running.
1046
1047	Frozen status of a cgroup doesn't affect any cgroup tree operations:
1048	it's possible to delete a frozen (and empty) cgroup, as well as
1049	create new sub-cgroups.
1050
1051  cgroup.kill
1052	A write-only single value file which exists in non-root cgroups.
1053	The only allowed value is "1".
1054
1055	Writing "1" to the file causes the cgroup and all descendant cgroups to
1056	be killed. This means that all processes located in the affected cgroup
1057	tree will be killed via SIGKILL.
1058
1059	Killing a cgroup tree will deal with concurrent forks appropriately and
1060	is protected against migrations.
1061
1062	In a threaded cgroup, writing this file fails with EOPNOTSUPP as
1063	killing cgroups is a process directed operation, i.e. it affects
1064	the whole thread-group.
1065
1066  cgroup.pressure
1067	A read-write single value file that allowed values are "0" and "1".
1068	The default is "1".
1069
1070	Writing "0" to the file will disable the cgroup PSI accounting.
1071	Writing "1" to the file will re-enable the cgroup PSI accounting.
1072
1073	This control attribute is not hierarchical, so disable or enable PSI
1074	accounting in a cgroup does not affect PSI accounting in descendants
1075	and doesn't need pass enablement via ancestors from root.
1076
1077	The reason this control attribute exists is that PSI accounts stalls for
1078	each cgroup separately and aggregates it at each level of the hierarchy.
1079	This may cause non-negligible overhead for some workloads when under
1080	deep level of the hierarchy, in which case this control attribute can
1081	be used to disable PSI accounting in the non-leaf cgroups.
1082
1083  irq.pressure
1084	A read-write nested-keyed file.
1085
1086	Shows pressure stall information for IRQ/SOFTIRQ. See
1087	:ref:`Documentation/accounting/psi.rst <psi>` for details.
1088
1089Controllers
1090===========
1091
1092.. _cgroup-v2-cpu:
1093
1094CPU
1095---
1096
1097The "cpu" controllers regulates distribution of CPU cycles.  This
1098controller implements weight and absolute bandwidth limit models for
1099normal scheduling policy and absolute bandwidth allocation model for
1100realtime scheduling policy.
1101
1102In all the above models, cycles distribution is defined only on a temporal
1103base and it does not account for the frequency at which tasks are executed.
1104The (optional) utilization clamping support allows to hint the schedutil
1105cpufreq governor about the minimum desired frequency which should always be
1106provided by a CPU, as well as the maximum desired frequency, which should not
1107be exceeded by a CPU.
1108
1109WARNING: cgroup2 cpu controller doesn't yet support the (bandwidth) control of
1110realtime processes. For a kernel built with the CONFIG_RT_GROUP_SCHED option
1111enabled for group scheduling of realtime processes, the cpu controller can only
1112be enabled when all RT processes are in the root cgroup. Be aware that system
1113management software may already have placed RT processes into non-root cgroups
1114during the system boot process, and these processes may need to be moved to the
1115root cgroup before the cpu controller can be enabled with a
1116CONFIG_RT_GROUP_SCHED enabled kernel.
1117
1118With CONFIG_RT_GROUP_SCHED disabled, this limitation does not apply and some of
1119the interface files either affect realtime processes or account for them. See
1120the following section for details. Only the cpu controller is affected by
1121CONFIG_RT_GROUP_SCHED. Other controllers can be used for the resource control of
1122realtime processes irrespective of CONFIG_RT_GROUP_SCHED.
1123
1124
1125CPU Interface Files
1126~~~~~~~~~~~~~~~~~~~
1127
1128The interaction of a process with the cpu controller depends on its scheduling
1129policy and the underlying scheduler. From the point of view of the cpu controller,
1130processes can be categorized as follows:
1131
1132* Processes under the fair-class scheduler
1133* Processes under a BPF scheduler with the corresponding ``cgroup_set_*`` callback
1134* Everything else: ``SCHED_{FIFO,RR,DEADLINE}`` and processes under a BPF scheduler
1135  without the corresponding ``cgroup_set_*`` callback
1136
1137For details on when a process is under the fair-class scheduler or a BPF scheduler,
1138check out :ref:`Documentation/scheduler/sched-ext.rst <sched-ext>`.
1139
1140For each of the following interface files, the above categories
1141will be referred to. All time durations are in microseconds.
1142
1143  cpu.stat
1144	A read-only flat-keyed file.
1145	This file exists whether the controller is enabled or not.
1146
1147	It always reports the following three stats, which account for all the
1148	processes in the cgroup (including those in descendant cgroups):
1149
1150	- usage_usec
1151	- user_usec
1152	- system_usec
1153
1154	and the following five when the controller is enabled, which account for
1155	only the processes under the fair-class scheduler:
1156
1157	- nr_periods
1158	- nr_throttled
1159	- throttled_usec
1160	- nr_bursts
1161	- burst_usec
1162
1163	Note that the above five CFS bandwidth stats are non-hierarchical;
1164	they only account for throttling caused by this cgroup's own bandwidth
1165	limit, not including throttling inherited from ancestor cgroups.
1166
1167  cpu.stat.local
1168	A read-only flat-keyed file.
1169	This file exists whether the controller is enabled or not.
1170
1171	It reports the following stat when the controller is enabled:
1172
1173	- throttled_usec
1174
1175	Unlike the ``throttled_usec`` reported by ``cpu.stat`` which
1176	accounts for throttling caused by this cgroup's own CFS
1177	bandwidth limit, ``cpu.stat.local`` reports the actual
1178	throttling time incurred by this cgroup's own runqueues,
1179	which may include throttling inherited from ancestor
1180	cgroup bandwidth limits.
1181
1182	When the controller is not enabled, this stat is not reported.
1183
1184  cpu.weight
1185	A read-write single value file which exists on non-root
1186	cgroups.  The default is "100".
1187
1188	For non idle groups (cpu.idle = 0), the weight is in the
1189	range [1, 10000].
1190
1191	If the cgroup has been configured to be SCHED_IDLE (cpu.idle = 1),
1192	then the weight will show as a 0.
1193
1194	This file affects only processes under the fair-class scheduler and a BPF
1195	scheduler with the ``cgroup_set_weight`` callback depending on what the
1196	callback actually does.
1197
1198  cpu.weight.nice
1199	A read-write single value file which exists on non-root
1200	cgroups.  The default is "0".
1201
1202	The nice value is in the range [-20, 19].
1203
1204	This interface file is an alternative interface for
1205	"cpu.weight" and allows reading and setting weight using the
1206	same values used by nice(2).  Because the range is smaller and
1207	granularity is coarser for the nice values, the read value is
1208	the closest approximation of the current weight.
1209
1210	This file affects only processes under the fair-class scheduler and a BPF
1211	scheduler with the ``cgroup_set_weight`` callback depending on what the
1212	callback actually does.
1213
1214  cpu.max
1215	A read-write two value file which exists on non-root cgroups.
1216	The default is "max 100000".
1217
1218	The maximum bandwidth limit.  It's in the following format::
1219
1220	  $MAX $PERIOD
1221
1222	which indicates that the group may consume up to $MAX in each
1223	$PERIOD duration.  "max" for $MAX indicates no limit.  If only
1224	one number is written, $MAX is updated.
1225
1226	This file affects only processes under the fair-class scheduler and a BPF
1227	scheduler with the ``cgroup_set_bandwidth`` callback depending on what
1228	the callback actually does.
1229
1230  cpu.max.burst
1231	A read-write single value file which exists on non-root
1232	cgroups.  The default is "0".
1233
1234	The burst in the range [0, $MAX].
1235
1236	This file affects only processes under the fair-class scheduler and a BPF
1237	scheduler with the ``cgroup_set_bandwidth`` callback depending on what
1238	the callback actually does.
1239
1240  cpu.pressure
1241	A read-write nested-keyed file.
1242
1243	Shows pressure stall information for CPU. See
1244	:ref:`Documentation/accounting/psi.rst <psi>` for details.
1245
1246	This file accounts for all the processes in the cgroup.
1247
1248  cpu.uclamp.min
1249	A read-write single value file which exists on non-root cgroups.
1250	The default is "0", i.e. no utilization boosting.
1251
1252	The requested minimum utilization (protection) as a percentage
1253	rational number, e.g. 12.34 for 12.34%.
1254
1255	This interface allows reading and setting minimum utilization clamp
1256	values similar to the sched_setattr(2). This minimum utilization
1257	value is used to clamp the task specific minimum utilization clamp,
1258	including those of realtime processes.
1259
1260	The requested minimum utilization (protection) is always capped by
1261	the current value for the maximum utilization (limit), i.e.
1262	`cpu.uclamp.max`.
1263
1264	This file affects all the processes in the cgroup.
1265
1266  cpu.uclamp.max
1267	A read-write single value file which exists on non-root cgroups.
1268	The default is "max". i.e. no utilization capping
1269
1270	The requested maximum utilization (limit) as a percentage rational
1271	number, e.g. 98.76 for 98.76%.
1272
1273	This interface allows reading and setting maximum utilization clamp
1274	values similar to the sched_setattr(2). This maximum utilization
1275	value is used to clamp the task specific maximum utilization clamp,
1276	including those of realtime processes.
1277
1278	This file affects all the processes in the cgroup.
1279
1280  cpu.idle
1281	A read-write single value file which exists on non-root cgroups.
1282	The default is 0.
1283
1284	This is the cgroup analog of the per-task SCHED_IDLE sched policy.
1285	Setting this value to a 1 will make the scheduling policy of the
1286	cgroup SCHED_IDLE. The threads inside the cgroup will retain their
1287	own relative priorities, but the cgroup itself will be treated as
1288	very low priority relative to its peers.
1289
1290	This file affects only processes under the fair-class scheduler and a BPF
1291	scheduler with the ``cgroup_set_idle`` callback depending on what the
1292	callback actually does.
1293
1294Memory
1295------
1296
1297The "memory" controller regulates distribution of memory.  Memory is
1298stateful and implements both limit and protection models.  Due to the
1299intertwining between memory usage and reclaim pressure and the
1300stateful nature of memory, the distribution model is relatively
1301complex.
1302
1303While not completely water-tight, all major memory usages by a given
1304cgroup are tracked so that the total memory consumption can be
1305accounted and controlled to a reasonable extent.  Currently, the
1306following types of memory usages are tracked.
1307
1308- Userland memory - page cache and anonymous memory.
1309
1310- Kernel data structures such as dentries and inodes.
1311
1312- TCP socket buffers.
1313
1314The above list may expand in the future for better coverage.
1315
1316
1317Memory Interface Files
1318~~~~~~~~~~~~~~~~~~~~~~
1319
1320All memory amounts are in bytes.  If a value which is not aligned to
1321PAGE_SIZE is written, the value may be rounded up to the closest
1322PAGE_SIZE multiple when read back.
1323
1324For the limit files described below, an empty or all-whitespace
1325write is accepted and sets the limit to 0.  To disable a limit,
1326write "max"; to set it to zero explicitly, write "0".
1327
1328  memory.current
1329	A read-only single value file which exists on non-root
1330	cgroups.
1331
1332	The total amount of memory currently being used by the cgroup
1333	and its descendants.
1334
1335  memory.min
1336	A read-write single value file which exists on non-root
1337	cgroups.  The default is "0".
1338
1339	Hard memory protection.  If the memory usage of a cgroup
1340	is within its effective min boundary, the cgroup's memory
1341	won't be reclaimed under any conditions. If there is no
1342	unprotected reclaimable memory available, OOM killer
1343	is invoked. Above the effective min boundary (or
1344	effective low boundary if it is higher), pages are reclaimed
1345	proportionally to the overage, reducing reclaim pressure for
1346	smaller overages.
1347
1348	Effective min boundary is limited by memory.min values of
1349	ancestor cgroups. If there is memory.min overcommitment
1350	(child cgroup or cgroups are requiring more protected memory
1351	than parent will allow), then each child cgroup will get
1352	the part of parent's protection proportional to its
1353	actual memory usage below memory.min.
1354
1355	Putting more memory than generally available under this
1356	protection is discouraged and may lead to constant OOMs.
1357
1358  memory.low
1359	A read-write single value file which exists on non-root
1360	cgroups.  The default is "0".
1361
1362	Best-effort memory protection.  If the memory usage of a
1363	cgroup is within its effective low boundary, the cgroup's
1364	memory won't be reclaimed unless there is no reclaimable
1365	memory available in unprotected cgroups.
1366	Above the effective low	boundary (or
1367	effective min boundary if it is higher), pages are reclaimed
1368	proportionally to the overage, reducing reclaim pressure for
1369	smaller overages.
1370
1371	Effective low boundary is limited by memory.low values of
1372	ancestor cgroups. If there is memory.low overcommitment
1373	(child cgroup or cgroups are requiring more protected memory
1374	than parent will allow), then each child cgroup will get
1375	the part of parent's protection proportional to its
1376	actual memory usage below memory.low.
1377
1378	Putting more memory than generally available under this
1379	protection is discouraged.
1380
1381  memory.high
1382	A read-write single value file which exists on non-root
1383	cgroups.  The default is "max".
1384
1385	Memory usage throttle limit.  If a cgroup's usage goes
1386	over the high boundary, the processes of the cgroup are
1387	throttled and put under heavy reclaim pressure.
1388
1389	Going over the high limit never invokes the OOM killer and
1390	under extreme conditions the limit may be breached. The high
1391	limit should be used in scenarios where an external process
1392	monitors the limited cgroup to alleviate heavy reclaim
1393	pressure.
1394
1395	If memory.high is opened with O_NONBLOCK then the synchronous
1396	reclaim is bypassed. This is useful for admin processes that
1397	need to dynamically adjust the job's memory limits without
1398	expending their own CPU resources on memory reclamation. The
1399	job will trigger the reclaim and/or get throttled on its
1400	next charge request.
1401
1402	Please note that with O_NONBLOCK, there is a chance that the
1403	target memory cgroup may take indefinite amount of time to
1404	reduce usage below the limit due to delayed charge request or
1405	busy-hitting its memory to slow down reclaim.
1406
1407  memory.max
1408	A read-write single value file which exists on non-root
1409	cgroups.  The default is "max".
1410
1411	Memory usage hard limit.  This is the main mechanism to limit
1412	memory usage of a cgroup.  If a cgroup's memory usage reaches
1413	this limit and can't be reduced, the OOM killer is invoked in
1414	the cgroup. Under certain circumstances, the usage may go
1415	over the limit temporarily.
1416
1417	In default configuration regular 0-order allocations always
1418	succeed unless OOM killer chooses current task as a victim.
1419
1420	Some kinds of allocations don't invoke the OOM killer.
1421	Caller could retry them differently, return into userspace
1422	as -ENOMEM or silently ignore in cases like disk readahead.
1423
1424	If memory.max is opened with O_NONBLOCK, then the synchronous
1425	reclaim and oom-kill are bypassed. This is useful for admin
1426	processes that need to dynamically adjust the job's memory limits
1427	without expending their own CPU resources on memory reclamation.
1428	The job will trigger the reclaim and/or oom-kill on its next
1429	charge request.
1430
1431	Please note that with O_NONBLOCK, there is a chance that the
1432	target memory cgroup may take indefinite amount of time to
1433	reduce usage below the limit due to delayed charge request or
1434	busy-hitting its memory to slow down reclaim.
1435
1436  memory.reclaim
1437	A write-only nested-keyed file which exists for all cgroups.
1438
1439	This is a simple interface to trigger memory reclaim in the
1440	target cgroup.
1441
1442	Example::
1443
1444	  echo "1G" > memory.reclaim
1445
1446	Please note that the kernel can over or under reclaim from
1447	the target cgroup. If less bytes are reclaimed than the
1448	specified amount, -EAGAIN is returned.
1449
1450	Please note that the proactive reclaim (triggered by this
1451	interface) is not meant to indicate memory pressure on the
1452	memory cgroup. Therefore socket memory balancing triggered by
1453	the memory reclaim normally is not exercised in this case.
1454	This means that the networking layer will not adapt based on
1455	reclaim induced by memory.reclaim.
1456
1457The following nested keys are defined.
1458
1459	  ==========            ================================
1460	  swappiness            Swappiness value to reclaim with
1461	  ==========            ================================
1462
1463	Specifying a swappiness value instructs the kernel to perform
1464	the reclaim with that swappiness value. Note that this has the
1465	same semantics as vm.swappiness applied to memcg reclaim with
1466	all the existing limitations and potential future extensions.
1467
1468	The valid range for swappiness is [0-200, max], setting
1469	swappiness=max exclusively reclaims anonymous memory.
1470
1471  memory.peak
1472	A read-write single value file which exists on non-root cgroups.
1473
1474	The max memory usage recorded for the cgroup and its descendants since
1475	either the creation of the cgroup or the most recent reset for that FD.
1476
1477	A write of any non-empty string to this file resets it to the
1478	current memory usage for subsequent reads through the same
1479	file descriptor.
1480
1481  memory.oom.group
1482	A read-write single value file which exists on non-root
1483	cgroups.  The default value is "0".
1484
1485	Determines whether the cgroup should be treated as
1486	an indivisible workload by the OOM killer. If set,
1487	all tasks belonging to the cgroup or to its descendants
1488	(if the memory cgroup is not a leaf cgroup) are killed
1489	together or not at all. This can be used to avoid
1490	partial kills to guarantee workload integrity.
1491
1492	Tasks with the OOM protection (oom_score_adj set to -1000)
1493	are treated as an exception and are never killed.
1494
1495	If the OOM killer is invoked in a cgroup, it's not going
1496	to kill any tasks outside of this cgroup, regardless
1497	memory.oom.group values of ancestor cgroups.
1498
1499  memory.events
1500	A read-only flat-keyed file which exists on non-root cgroups.
1501	The following entries are defined.  Unless specified
1502	otherwise, a value change in this file generates a file
1503	modified event.
1504
1505	Note that all fields in this file are hierarchical and the
1506	file modified event can be generated due to an event down the
1507	hierarchy. For the local events at the cgroup level see
1508	memory.events.local.
1509
1510	  low
1511		The number of times the cgroup is reclaimed due to
1512		high memory pressure even though its usage is under
1513		the low boundary.  This usually indicates that the low
1514		boundary is over-committed.
1515
1516	  high
1517		The number of times processes of the cgroup are
1518		throttled and routed to perform direct memory reclaim
1519		because the high memory boundary was exceeded.  For a
1520		cgroup whose memory usage is capped by the high limit
1521		rather than global memory pressure, this event's
1522		occurrences are expected.
1523
1524	  max
1525		The number of times the cgroup's memory usage was
1526		about to go over the max boundary.  If direct reclaim
1527		fails to bring it down, the cgroup goes to OOM state.
1528
1529	  oom
1530		The number of time the cgroup's memory usage was
1531		reached the limit and allocation was about to fail.
1532
1533		This event is not raised if the OOM killer is not
1534		considered as an option, e.g. for failed high-order
1535		allocations or if caller asked to not retry attempts.
1536
1537	  oom_kill
1538		The number of processes belonging to this cgroup
1539		killed by any kind of OOM killer.
1540
1541          oom_group_kill
1542                The number of times a group OOM has occurred.
1543
1544          sock_throttled
1545                The number of times network sockets associated with
1546                this cgroup are throttled.
1547
1548  memory.events.local
1549	Similar to memory.events but the fields in the file are local
1550	to the cgroup i.e. not hierarchical. The file modified event
1551	generated on this file reflects only the local events.
1552
1553  memory.stat
1554	A read-only flat-keyed file which exists on non-root cgroups.
1555
1556	This breaks down the cgroup's memory footprint into different
1557	types of memory, type-specific details, and other information
1558	on the state and past events of the memory management system.
1559
1560	All memory amounts are in bytes.
1561
1562	The entries are ordered to be human readable, and new entries
1563	can show up in the middle. Don't rely on items remaining in a
1564	fixed position; use the keys to look up specific values!
1565
1566	If the entry has no per-node counter (or not show in the
1567	memory.numa_stat). We use 'npn' (non-per-node) as the tag
1568	to indicate that it will not show in the memory.numa_stat.
1569
1570	  anon
1571		Amount of memory used in anonymous mappings such as
1572		brk(), sbrk(), and mmap(MAP_ANONYMOUS). Note that
1573		some kernel configurations might account complete larger
1574		allocations (e.g., THP) if only some, but not all the
1575		memory of such an allocation is mapped anymore.
1576
1577	  file
1578		Amount of memory used to cache filesystem data,
1579		including tmpfs and shared memory.
1580
1581	  kernel (npn)
1582		Amount of total kernel memory, including
1583		(kernel_stack, pagetables, percpu, vmalloc, slab) in
1584		addition to other kernel memory use cases.
1585
1586	  kernel_stack
1587		Amount of memory allocated to kernel stacks.
1588
1589	  pagetables
1590                Amount of memory allocated for page tables.
1591
1592	  sec_pagetables
1593		Amount of memory allocated for secondary page tables,
1594		this currently includes KVM mmu allocations on x86
1595		and arm64 and IOMMU page tables.
1596
1597	  percpu (npn)
1598		Amount of memory used for storing per-cpu kernel
1599		data structures.
1600
1601	  sock (npn)
1602		Amount of memory used in network transmission buffers
1603
1604	  vmalloc
1605		Amount of memory used for vmap backed memory.
1606
1607	  shmem
1608		Amount of cached filesystem data that is swap-backed,
1609		such as tmpfs, shm segments, shared anonymous mmap()s
1610
1611	  zswap
1612		Amount of memory consumed by the zswap compression backend.
1613
1614	  zswapped
1615		Amount of application memory swapped out to zswap.
1616
1617	  file_mapped
1618		Amount of cached filesystem data mapped with mmap(). Note
1619		that some kernel configurations might account complete
1620		larger allocations (e.g., THP) if only some, but not
1621		not all the memory of such an allocation is mapped.
1622
1623	  file_dirty
1624		Amount of cached filesystem data that was modified but
1625		not yet written back to disk
1626
1627	  file_writeback
1628		Amount of cached filesystem data that was modified and
1629		is currently being written back to disk
1630
1631	  swapcached
1632		Amount of swap cached in memory. The swapcache is accounted
1633		against both memory and swap usage.
1634
1635	  anon_thp
1636		Amount of memory used in anonymous mappings backed by
1637		transparent hugepages
1638
1639	  file_thp
1640		Amount of cached filesystem data backed by transparent
1641		hugepages
1642
1643	  shmem_thp
1644		Amount of shm, tmpfs, shared anonymous mmap()s backed by
1645		transparent hugepages
1646
1647	  inactive_anon, active_anon, inactive_file, active_file, unevictable
1648		Amount of memory, swap-backed and filesystem-backed,
1649		on the internal memory management lists used by the
1650		page reclaim algorithm.
1651
1652		As these represent internal list state (eg. shmem pages are on anon
1653		memory management lists), inactive_foo + active_foo may not be equal to
1654		the value for the foo counter, since the foo counter is type-based, not
1655		list-based.
1656
1657	  slab_reclaimable
1658		Part of "slab" that might be reclaimed, such as
1659		dentries and inodes.
1660
1661	  slab_unreclaimable
1662		Part of "slab" that cannot be reclaimed on memory
1663		pressure.
1664
1665	  slab (npn)
1666		Amount of memory used for storing in-kernel data
1667		structures.
1668
1669	  workingset_refault_anon
1670		Number of refaults of previously evicted anonymous pages.
1671
1672	  workingset_refault_file
1673		Number of refaults of previously evicted file pages.
1674
1675	  workingset_activate_anon
1676		Number of refaulted anonymous pages that were immediately
1677		activated.
1678
1679	  workingset_activate_file
1680		Number of refaulted file pages that were immediately activated.
1681
1682	  workingset_restore_anon
1683		Number of restored anonymous pages which have been detected as
1684		an active workingset before they got reclaimed.
1685
1686	  workingset_restore_file
1687		Number of restored file pages which have been detected as an
1688		active workingset before they got reclaimed.
1689
1690	  workingset_nodereclaim
1691		Number of times a shadow node has been reclaimed
1692
1693	  pswpin (npn)
1694		Number of pages swapped into memory
1695
1696	  pswpout (npn)
1697		Number of pages swapped out of memory
1698
1699	  pgscan (npn)
1700		Amount of scanned pages (in an inactive LRU list)
1701
1702	  pgsteal (npn)
1703		Amount of reclaimed pages
1704
1705	  pgscan_kswapd (npn)
1706		Amount of scanned pages by kswapd (in an inactive LRU list)
1707
1708	  pgscan_direct (npn)
1709		Amount of scanned pages directly  (in an inactive LRU list)
1710
1711	  pgscan_khugepaged (npn)
1712		Amount of scanned pages by khugepaged  (in an inactive LRU list)
1713
1714	  pgscan_proactive (npn)
1715		Amount of scanned pages proactively (in an inactive LRU list)
1716
1717	  pgsteal_kswapd (npn)
1718		Amount of reclaimed pages by kswapd
1719
1720	  pgsteal_direct (npn)
1721		Amount of reclaimed pages directly
1722
1723	  pgsteal_khugepaged (npn)
1724		Amount of reclaimed pages by khugepaged
1725
1726	  pgsteal_proactive (npn)
1727		Amount of reclaimed pages proactively
1728
1729	  pgfault (npn)
1730		Total number of page faults incurred
1731
1732	  pgmajfault (npn)
1733		Number of major page faults incurred
1734
1735	  pgrefill (npn)
1736		Amount of scanned pages (in an active LRU list)
1737
1738	  pgactivate (npn)
1739		Amount of pages moved to the active LRU list
1740
1741	  pgdeactivate (npn)
1742		Amount of pages moved to the inactive LRU list
1743
1744	  pglazyfree (npn)
1745		Amount of pages postponed to be freed under memory pressure
1746
1747	  pglazyfreed (npn)
1748		Amount of reclaimed lazyfree pages
1749
1750	  swpin_zero
1751		Number of pages swapped into memory and filled with zero, where I/O
1752		was optimized out because the page content was detected to be zero
1753		during swapout.
1754
1755	  swpout_zero
1756		Number of zero-filled pages swapped out with I/O skipped due to the
1757		content being detected as zero.
1758
1759	  zswpin
1760		Number of pages moved in to memory from zswap.
1761
1762	  zswpout
1763		Number of pages moved out of memory to zswap.
1764
1765	  zswpwb
1766		Number of pages written from zswap to swap.
1767
1768	  zswap_incomp
1769		Amount of memory used by incompressible pages currently stored in zswap
1770		without compression. These pages could not be compressed to
1771		a size smaller than PAGE_SIZE, so they are stored as-is.
1772
1773	  thp_fault_alloc (npn)
1774		Number of transparent hugepages which were allocated to satisfy
1775		a page fault. This counter is not present when CONFIG_TRANSPARENT_HUGEPAGE
1776                is not set.
1777
1778	  thp_collapse_alloc (npn)
1779		Number of transparent hugepages which were allocated to allow
1780		collapsing an existing range of pages. This counter is not
1781		present when CONFIG_TRANSPARENT_HUGEPAGE is not set.
1782
1783	  thp_swpout (npn)
1784		Number of transparent hugepages which are swapout in one piece
1785		without splitting.
1786
1787	  thp_swpout_fallback (npn)
1788		Number of transparent hugepages which were split before swapout.
1789		Usually because failed to allocate some continuous swap space
1790		for the huge page.
1791
1792	  numa_pages_migrated (npn)
1793		Number of pages migrated by NUMA balancing.
1794
1795	  numa_pte_updates (npn)
1796		Number of pages whose page table entries are modified by
1797		NUMA balancing to produce NUMA hinting faults on access.
1798
1799	  numa_hint_faults (npn)
1800		Number of NUMA hinting faults.
1801
1802	  pgdemote_kswapd
1803		Number of pages demoted by kswapd.
1804
1805	  pgdemote_direct
1806		Number of pages demoted directly.
1807
1808	  pgdemote_khugepaged
1809		Number of pages demoted by khugepaged.
1810
1811	  pgdemote_proactive
1812		Number of pages demoted by proactively.
1813
1814	  hugetlb
1815		Amount of memory used by hugetlb pages. This metric only shows
1816		up if hugetlb usage is accounted for in memory.current (i.e.
1817		cgroup is mounted with the memory_hugetlb_accounting option).
1818
1819  memory.numa_stat
1820	A read-only nested-keyed file which exists on non-root cgroups.
1821
1822	This breaks down the cgroup's memory footprint into different
1823	types of memory, type-specific details, and other information
1824	per node on the state of the memory management system.
1825
1826	This is useful for providing visibility into the NUMA locality
1827	information within an memcg since the pages are allowed to be
1828	allocated from any physical node. One of the use case is evaluating
1829	application performance by combining this information with the
1830	application's CPU allocation.
1831
1832	All memory amounts are in bytes.
1833
1834	The output format of memory.numa_stat is::
1835
1836	  type N0=<bytes in node 0> N1=<bytes in node 1> ...
1837
1838	The entries are ordered to be human readable, and new entries
1839	can show up in the middle. Don't rely on items remaining in a
1840	fixed position; use the keys to look up specific values!
1841
1842	The entries can refer to the memory.stat.
1843
1844  memory.swap.current
1845	A read-only single value file which exists on non-root
1846	cgroups.
1847
1848	The total amount of swap currently being used by the cgroup
1849	and its descendants.
1850
1851  memory.swap.high
1852	A read-write single value file which exists on non-root
1853	cgroups.  The default is "max".
1854
1855	Swap usage throttle limit.  If a cgroup's swap usage exceeds
1856	this limit, all its further allocations will be throttled to
1857	allow userspace to implement custom out-of-memory procedures.
1858
1859	This limit marks a point of no return for the cgroup. It is NOT
1860	designed to manage the amount of swapping a workload does
1861	during regular operation. Compare to memory.swap.max, which
1862	prohibits swapping past a set amount, but lets the cgroup
1863	continue unimpeded as long as other memory can be reclaimed.
1864
1865	Healthy workloads are not expected to reach this limit.
1866
1867  memory.swap.peak
1868	A read-write single value file which exists on non-root cgroups.
1869
1870	The max swap usage recorded for the cgroup and its descendants since
1871	the creation of the cgroup or the most recent reset for that FD.
1872
1873	A write of any non-empty string to this file resets it to the
1874	current memory usage for subsequent reads through the same
1875	file descriptor.
1876
1877  memory.swap.max
1878	A read-write single value file which exists on non-root
1879	cgroups.  The default is "max".
1880
1881	Swap usage hard limit.  If a cgroup's swap usage reaches this
1882	limit, anonymous memory of the cgroup will not be swapped out.
1883
1884  memory.swap.events
1885	A read-only flat-keyed file which exists on non-root cgroups.
1886	The following entries are defined.  Unless specified
1887	otherwise, a value change in this file generates a file
1888	modified event.
1889
1890	  high
1891		The number of times the cgroup's swap usage was over
1892		the high threshold.
1893
1894	  max
1895		The number of times the cgroup's swap usage was about
1896		to go over the max boundary and swap allocation
1897		failed.
1898
1899	  fail
1900		The number of times swap allocation failed either
1901		because of running out of swap system-wide or max
1902		limit.
1903
1904	When reduced under the current usage, the existing swap
1905	entries are reclaimed gradually and the swap usage may stay
1906	higher than the limit for an extended period of time.  This
1907	reduces the impact on the workload and memory management.
1908
1909  memory.zswap.current
1910	A read-only single value file which exists on non-root
1911	cgroups.
1912
1913	The total amount of memory consumed by the zswap compression
1914	backend.
1915
1916  memory.zswap.max
1917	A read-write single value file which exists on non-root
1918	cgroups.  The default is "max".
1919
1920	Zswap usage hard limit. If a cgroup's zswap pool reaches this
1921	limit, it will refuse to take any more stores before existing
1922	entries fault back in or are written out to disk.
1923
1924  memory.zswap.writeback
1925	A read-write single value file. The default value is "1".
1926	Note that this setting is hierarchical, i.e. the writeback would be
1927	implicitly disabled for child cgroups if the upper hierarchy
1928	does so.
1929
1930	When this is set to 0, all swapping attempts to swapping devices
1931	are disabled. This included both zswap writebacks, and swapping due
1932	to zswap store failures. If the zswap store failures are recurring
1933	(for e.g if the pages are incompressible), users can observe
1934	reclaim inefficiency after disabling writeback (because the same
1935	pages might be rejected again and again).
1936
1937	Note that this is subtly different from setting memory.swap.max to
1938	0, as it still allows for pages to be written to the zswap pool.
1939	This setting has no effect if zswap is disabled, and swapping
1940	is allowed unless memory.swap.max is set to 0.
1941
1942  memory.pressure
1943	A read-write nested-keyed file.
1944
1945	Shows pressure stall information for memory. See
1946	:ref:`Documentation/accounting/psi.rst <psi>` for details.
1947
1948
1949Usage Guidelines
1950~~~~~~~~~~~~~~~~
1951
1952"memory.high" is the main mechanism to control memory usage.
1953Over-committing on high limit (sum of high limits > available memory)
1954and letting global memory pressure to distribute memory according to
1955usage is a viable strategy.
1956
1957Because breach of the high limit doesn't trigger the OOM killer but
1958throttles the offending cgroup, a management agent has ample
1959opportunities to monitor and take appropriate actions such as granting
1960more memory or terminating the workload.
1961
1962Determining whether a cgroup has enough memory is not trivial as
1963memory usage doesn't indicate whether the workload can benefit from
1964more memory.  For example, a workload which writes data received from
1965network to a file can use all available memory but can also operate as
1966performant with a small amount of memory.  A measure of memory
1967pressure - how much the workload is being impacted due to lack of
1968memory - is necessary to determine whether a workload needs more
1969memory; unfortunately, memory pressure monitoring mechanism isn't
1970implemented yet.
1971
1972Reclaim Protection
1973~~~~~~~~~~~~~~~~~~
1974
1975The protection configured with "memory.low" or "memory.min" applies relatively
1976to the target of the reclaim (i.e. any of memory cgroup limits, proactive
1977memory.reclaim or global reclaim apparently located in the root cgroup).
1978The protection value configured for B applies unchanged to the reclaim
1979targeting A (i.e. caused by competition with the sibling E)::
1980
1981		root - ... - A - B - C
1982		              \    ` D
1983		               ` E
1984
1985When the reclaim targets ancestors of A, the effective protection of B is
1986capped by the protection value configured for A (and any other intermediate
1987ancestors between A and the target).
1988
1989To express indifference about relative sibling protection, it is suggested to
1990use memory_recursiveprot. Configuring all descendants of a parent with finite
1991protection to "max" works but it may unnecessarily skew memory.events:low
1992field.
1993
1994Memory Ownership
1995~~~~~~~~~~~~~~~~
1996
1997A memory area is charged to the cgroup which instantiated it and stays
1998charged to the cgroup until the area is released.  Migrating a process
1999to a different cgroup doesn't move the memory usages that it
2000instantiated while in the previous cgroup to the new cgroup.
2001
2002A memory area may be used by processes belonging to different cgroups.
2003To which cgroup the area will be charged is in-deterministic; however,
2004over time, the memory area is likely to end up in a cgroup which has
2005enough memory allowance to avoid high reclaim pressure.
2006
2007If a cgroup sweeps a considerable amount of memory which is expected
2008to be accessed repeatedly by other cgroups, it may make sense to use
2009POSIX_FADV_DONTNEED to relinquish the ownership of memory areas
2010belonging to the affected files to ensure correct memory ownership.
2011
2012
2013IO
2014--
2015
2016The "io" controller regulates the distribution of IO resources.  This
2017controller implements both weight based and absolute bandwidth or IOPS
2018limit distribution.  Absolute BPS and IOPS limits are enforced by
2019blk-throttle and apply to all devices, while weight based proportional
2020distribution is provided by the iocost cost model controller
2021(CONFIG_BLK_CGROUP_IOCOST) and, when the BFQ I/O scheduler is in use
2022for a device, by BFQ's own cgroup support.  Latency-based protection
2023(CONFIG_BLK_CGROUP_IOLATENCY) and I/O priority assignment
2024(CONFIG_BLK_CGROUP_IOPRIO) are also available.
2025
2026
2027IO Interface Files
2028~~~~~~~~~~~~~~~~~~
2029
2030  io.stat
2031	A read-only nested-keyed file.
2032
2033	Lines are keyed by $MAJ:$MIN device numbers and not ordered.
2034	The following nested keys are defined.
2035
2036	  ======	=====================
2037	  rbytes	Bytes read
2038	  wbytes	Bytes written
2039	  rios		Number of read IOs
2040	  wios		Number of write IOs
2041	  dbytes	Bytes discarded
2042	  dios		Number of discard IOs
2043	  ======	=====================
2044
2045	An example read output follows::
2046
2047	  8:16 rbytes=1459200 wbytes=314773504 rios=192 wios=353 dbytes=0 dios=0
2048	  8:0 rbytes=90430464 wbytes=299008000 rios=8950 wios=1252 dbytes=50331648 dios=3021
2049
2050  io.cost.qos
2051	A read-write nested-keyed file which exists only on the root
2052	cgroup.
2053
2054	This file configures the Quality of Service of the IO cost
2055	model based controller (CONFIG_BLK_CGROUP_IOCOST) which
2056	currently implements "io.weight" proportional control.  Lines
2057	are keyed by $MAJ:$MIN device numbers and not ordered.  The
2058	line for a given device is populated on the first write for
2059	the device on "io.cost.qos" or "io.cost.model".  The following
2060	nested keys are defined.
2061
2062	  ======	=====================================
2063	  enable	Weight-based control enable
2064	  ctrl		"auto" or "user"
2065	  rpct		Read latency percentile    [0, 100]
2066	  rlat		Read latency threshold
2067	  wpct		Write latency percentile   [0, 100]
2068	  wlat		Write latency threshold
2069	  min		Minimum scaling percentage [1, 10000]
2070	  max		Maximum scaling percentage [1, 10000]
2071	  ======	=====================================
2072
2073	The controller is disabled by default and can be enabled by
2074	setting "enable" to 1.  "rpct" and "wpct" parameters default
2075	to zero and the controller uses internal device saturation
2076	state to adjust the overall IO rate between "min" and "max".
2077
2078	When a better control quality is needed, latency QoS
2079	parameters can be configured.  For example::
2080
2081	  8:16 enable=1 ctrl=auto rpct=95.00 rlat=75000 wpct=95.00 wlat=150000 min=50.00 max=150.0
2082
2083	shows that on sdb, the controller is enabled, will consider
2084	the device saturated if the 95th percentile of read completion
2085	latencies is above 75ms or write 150ms, and adjust the overall
2086	IO issue rate between 50% and 150% accordingly.
2087
2088	The lower the saturation point, the better the latency QoS at
2089	the cost of aggregate bandwidth.  The narrower the allowed
2090	adjustment range between "min" and "max", the more conformant
2091	to the cost model the IO behavior.  Note that the IO issue
2092	base rate may be far off from 100% and setting "min" and "max"
2093	blindly can lead to a significant loss of device capacity or
2094	control quality.  "min" and "max" are useful for regulating
2095	devices which show wide temporary behavior changes - e.g. a
2096	ssd which accepts writes at the line speed for a while and
2097	then completely stalls for multiple seconds.
2098
2099	When "ctrl" is "auto", the parameters are controlled by the
2100	kernel and may change automatically.  Setting "ctrl" to "user"
2101	or setting any of the percentile and latency parameters puts
2102	it into "user" mode and disables the automatic changes.  The
2103	automatic mode can be restored by setting "ctrl" to "auto".
2104
2105  io.cost.model
2106	A read-write nested-keyed file which exists only on the root
2107	cgroup.
2108
2109	This file configures the cost model of the IO cost model based
2110	controller (CONFIG_BLK_CGROUP_IOCOST) which currently
2111	implements "io.weight" proportional control.  Lines are keyed
2112	by $MAJ:$MIN device numbers and not ordered.  The line for a
2113	given device is populated on the first write for the device on
2114	"io.cost.qos" or "io.cost.model".  The following nested keys
2115	are defined.
2116
2117	  =====		================================
2118	  ctrl		"auto" or "user"
2119	  model		The cost model in use - "linear"
2120	  =====		================================
2121
2122	When "ctrl" is "auto", the kernel may change all parameters
2123	dynamically.  When "ctrl" is set to "user" or any other
2124	parameters are written to, "ctrl" become "user" and the
2125	automatic changes are disabled.
2126
2127	When "model" is "linear", the following model parameters are
2128	defined.
2129
2130	  =============	========================================
2131	  [r|w]bps	The maximum sequential IO throughput
2132	  [r|w]seqiops	The maximum 4k sequential IOs per second
2133	  [r|w]randiops	The maximum 4k random IOs per second
2134	  flushiops	The maximum flushes per second
2135	  =============	========================================
2136
2137	From the above, the builtin linear model determines the base
2138	costs of a sequential and random IO and the cost coefficient
2139	for the IO size.  While simple, this model can cover most
2140	common device classes acceptably.
2141
2142	"flushiops" determines the cost of a cache flush: a write with
2143	a preceding cache flush is charged one flush on top of its data
2144	cost, and a FUA write one more flush on devices without native
2145	FUA support.  It is zero in the builtin profiles, so flushes
2146	stay free until it is configured.
2147
2148	The IO cost model isn't expected to be accurate in absolute
2149	sense and is scaled to the device behavior dynamically.
2150
2151	If needed, tools/cgroup/iocost_coef_gen.py can be used to
2152	generate device-specific coefficients.
2153
2154  io.weight
2155	A read-write flat-keyed file which exists on non-root cgroups.
2156	The default is "default 100".
2157
2158	The first line is the default weight applied to devices
2159	without specific override.  The rest are overrides keyed by
2160	$MAJ:$MIN device numbers and not ordered.  The weights are in
2161	the range [1, 10000] and specifies the relative amount IO time
2162	the cgroup can use in relation to its siblings.
2163
2164	The default weight can be updated by writing either "default
2165	$WEIGHT" or simply "$WEIGHT".  Overrides can be set by writing
2166	"$MAJ:$MIN $WEIGHT" and unset by writing "$MAJ:$MIN default".
2167
2168	An example read output follows::
2169
2170	  default 100
2171	  8:16 200
2172	  8:0 50
2173
2174  io.max
2175	A read-write nested-keyed file which exists on non-root
2176	cgroups.
2177
2178	BPS and IOPS based IO limit.  Lines are keyed by $MAJ:$MIN
2179	device numbers and not ordered.  The following nested keys are
2180	defined.
2181
2182	  =====		==================================
2183	  rbps		Max read bytes per second
2184	  wbps		Max write bytes per second
2185	  riops		Max read IO operations per second
2186	  wiops		Max write IO operations per second
2187	  =====		==================================
2188
2189	When writing, any number of nested key-value pairs can be
2190	specified in any order.  "max" can be specified as the value
2191	to remove a specific limit.  If the same key is specified
2192	multiple times, the outcome is undefined.
2193
2194	BPS and IOPS are measured in each IO direction and IOs are
2195	delayed if limit is reached.  Temporary bursts are allowed.
2196
2197	Setting read limit at 2M BPS and write at 120 IOPS for 8:16::
2198
2199	  echo "8:16 rbps=2097152 wiops=120" > io.max
2200
2201	Reading returns the following::
2202
2203	  8:16 rbps=2097152 wbps=max riops=max wiops=120
2204
2205	Write IOPS limit can be removed by writing the following::
2206
2207	  echo "8:16 wiops=max" > io.max
2208
2209	Reading now returns the following::
2210
2211	  8:16 rbps=2097152 wbps=max riops=max wiops=max
2212
2213  io.pressure
2214	A read-write nested-keyed file.
2215
2216	Shows pressure stall information for IO. See
2217	:ref:`Documentation/accounting/psi.rst <psi>` for details.
2218
2219
2220Writeback
2221~~~~~~~~~
2222
2223Page cache is dirtied through buffered writes and shared mmaps and
2224written asynchronously to the backing filesystem by the writeback
2225mechanism.  Writeback sits between the memory and IO domains and
2226regulates the proportion of dirty memory by balancing dirtying and
2227write IOs.
2228
2229The io controller, in conjunction with the memory controller,
2230implements control of page cache writeback IOs.  The memory controller
2231defines the memory domain that dirty memory ratio is calculated and
2232maintained for and the io controller defines the io domain which
2233writes out dirty pages for the memory domain.  Both system-wide and
2234per-cgroup dirty memory states are examined and the more restrictive
2235of the two is enforced.
2236
2237cgroup writeback requires explicit support from the underlying
2238filesystem.  Currently, cgroup writeback is implemented on ext2, ext4,
2239btrfs, f2fs, and xfs.  On other filesystems, all writeback IOs are
2240attributed to the root cgroup.
2241
2242There are inherent differences in memory and writeback management
2243which affects how cgroup ownership is tracked.  Memory is tracked per
2244page while writeback per inode.  For the purpose of writeback, an
2245inode is assigned to a cgroup and all IO requests to write dirty pages
2246from the inode are attributed to that cgroup.
2247
2248As cgroup ownership for memory is tracked per page, there can be pages
2249which are associated with different cgroups than the one the inode is
2250associated with.  These are called foreign pages.  The writeback
2251constantly keeps track of foreign pages and, if a particular foreign
2252cgroup becomes the majority over a certain period of time, switches
2253the ownership of the inode to that cgroup.
2254
2255While this model is enough for most use cases where a given inode is
2256mostly dirtied by a single cgroup even when the main writing cgroup
2257changes over time, use cases where multiple cgroups write to a single
2258inode simultaneously are not supported well.  In such circumstances, a
2259significant portion of IOs are likely to be attributed incorrectly.
2260As memory controller assigns page ownership on the first use and
2261doesn't update it until the page is released, even if writeback
2262strictly follows page ownership, multiple cgroups dirtying overlapping
2263areas wouldn't work as expected.  It's recommended to avoid such usage
2264patterns.
2265
2266The sysctl knobs which affect writeback behavior are applied to cgroup
2267writeback as follows.
2268
2269  vm.dirty_background_ratio, vm.dirty_ratio
2270	These ratios apply the same to cgroup writeback with the
2271	amount of available memory capped by limits imposed by the
2272	memory controller and system-wide clean memory.
2273
2274  vm.dirty_background_bytes, vm.dirty_bytes
2275	For cgroup writeback, this is calculated into ratio against
2276	total available memory and applied the same way as
2277	vm.dirty[_background]_ratio.
2278
2279
2280IO Latency
2281~~~~~~~~~~
2282
2283This is a cgroup v2 controller for IO workload protection.  You provide a group
2284with a latency target, and if the group misses its target the controller will
2285throttle any peers that have a lower latency target than the protected
2286workload.  How a miss is detected depends on the device: on rotational devices
2287the average latency over the window must exceed the target, while on
2288non-rotational devices a miss is counted once enough of the IOs in the window
2289individually exceed the target.
2290
2291The limits are only applied at the peer level in the hierarchy.  This means that
2292in the diagram below, only groups A, B, and C will influence each other, and
2293groups D and F will influence each other.  Group G will influence nobody::
2294
2295			[root]
2296		/	   |		\
2297		A	   B		C
2298	       /  \        |
2299	      D    F	   G
2300
2301
2302So the ideal way to configure this is to set io.latency in groups A, B, and C.
2303Generally you do not want to set a value lower than the latency your device
2304supports.  Experiment to find the value that works best for your workload.
2305Start at higher than the expected latency for your device and, with
2306blkcg_debug_stats enabled, observe io.stat for your workload group to get an
2307idea of the latency you see during normal operation.  On rotational devices,
2308use the avg_lat value as a basis for your real setting, setting it 10-15%
2309higher.  On non-rotational devices io.stat reports no average latency; set
2310the target based on your device and use the missed/total fields to verify it
2311is being met.
2312
2313How IO Latency Throttling Works
2314~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2315
2316io.latency is work conserving; so as long as everybody is meeting their latency
2317target the controller doesn't do anything.  Once a group starts missing its
2318target it begins throttling any peer group that has a higher target than itself.
2319This throttling takes 2 forms:
2320
2321- Queue depth throttling.  This is the number of outstanding IO's a group is
2322  allowed to have.  We will clamp down relatively quickly, starting at no limit
2323  and going all the way down to 1 IO at a time.
2324
2325- Artificial delay induction.  There are certain types of IO that cannot be
2326  throttled without possibly adversely affecting higher priority groups.  This
2327  includes swapping and metadata IO.  These types of IO are allowed to occur
2328  normally, however they are "charged" to the originating group.  If the
2329  originating group is being throttled you will see the use_delay and delay_nsec
2330  fields in io.stat increase.  The delay_nsec value is how many nanoseconds that
2331  are being added to any process that runs in this group.  Because this number can
2332  grow quite large if there is a lot of swapping or metadata IO occurring we
2333  limit the individual delay events to 1 second at a time.
2334
2335Once the victimized group starts meeting its latency target again it will start
2336unthrottling any peer groups that were throttled previously.  If the victimized
2337group simply stops doing IO the global counter will unthrottle appropriately.
2338
2339IO Latency Interface Files
2340~~~~~~~~~~~~~~~~~~~~~~~~~~
2341
2342  io.latency
2343	This takes a similar format as the other controllers.
2344
2345		"MAJOR:MINOR target=<target time in microseconds>"
2346
2347  io.stat
2348	If the controller is enabled you will see extra stats in io.stat in
2349	addition to the normal ones.  These debug stats are only emitted when
2350	the blkcg_debug_stats module parameter is enabled (it is disabled by
2351	default).
2352
2353	The reported latency fields depend on the device.  Rotational devices
2354	report avg_lat and win; non-rotational devices report missed and total
2355	instead.  missed and total are live counters for the current window and
2356	may change between reads.
2357
2358	  depth
2359		This is the current queue depth for the group.
2360
2361	  avg_lat
2362		(Rotational devices only.)  This is an exponential moving
2363		average with a decay rate of 1/exp bound by the sampling
2364		interval.  The decay rate interval can be calculated by
2365		multiplying the win value in io.stat by the corresponding number
2366		of samples based on the win value.
2367
2368	  win
2369		(Rotational devices only.)  The sampling window size in
2370		milliseconds.  This is the minimum duration of time between
2371		evaluation events.  Windows only elapse with IO activity.  Idle
2372		periods extend the most recent window.
2373
2374	  missed
2375		(Non-rotational devices only.)  The number of IOs in the
2376		current window whose latency exceeded the target.  A group is
2377		considered to be missing its target once missed reaches a
2378		certain ratio of total.
2379
2380	  total
2381		(Non-rotational devices only.)  The total number of IOs
2382		accounted in the current window.
2383
2384IO Priority
2385~~~~~~~~~~~
2386
2387A single attribute controls the behavior of the I/O priority cgroup policy,
2388namely the io.prio.class attribute. The following values are accepted for
2389that attribute:
2390
2391  no-change
2392	Do not modify the I/O priority class.
2393
2394  promote-to-rt
2395	For requests that have a non-RT I/O priority class, change it into RT.
2396	Also change the priority level of these requests to 4. Do not modify
2397	the I/O priority of requests that have priority class RT.
2398
2399  restrict-to-be
2400	For requests that do not have an I/O priority class or that have I/O
2401	priority class RT, change it into BE. Also change the priority level
2402	of these requests to 0. Do not modify the I/O priority class of
2403	requests that have priority class IDLE.
2404
2405  idle
2406	Change the I/O priority class of all requests into IDLE, the lowest
2407	I/O priority class.
2408
2409  none-to-rt
2410	Deprecated. Just an alias for promote-to-rt.
2411
2412The following numerical values are associated with the I/O priority policies:
2413
2414+----------------+---+
2415| no-change      | 0 |
2416+----------------+---+
2417| promote-to-rt  | 1 |
2418+----------------+---+
2419| restrict-to-be | 2 |
2420+----------------+---+
2421| idle           | 3 |
2422+----------------+---+
2423
2424The numerical value that corresponds to each I/O priority class is as follows:
2425
2426+-------------------------------+---+
2427| IOPRIO_CLASS_NONE             | 0 |
2428+-------------------------------+---+
2429| IOPRIO_CLASS_RT (real-time)   | 1 |
2430+-------------------------------+---+
2431| IOPRIO_CLASS_BE (best effort) | 2 |
2432+-------------------------------+---+
2433| IOPRIO_CLASS_IDLE             | 3 |
2434+-------------------------------+---+
2435
2436The algorithm to set the I/O priority class for a request is as follows:
2437
2438- If I/O priority class policy is promote-to-rt, change the request I/O
2439  priority class to IOPRIO_CLASS_RT and change the request I/O priority
2440  level to 4.
2441- If I/O priority class policy is not promote-to-rt, translate the I/O priority
2442  class policy into a number, then change the request I/O priority class
2443  into the maximum of the I/O priority class policy number and the numerical
2444  I/O priority class.
2445
2446PID
2447---
2448
2449The process number controller is used to allow a cgroup to stop any
2450new tasks from being fork()'d or clone()'d after a specified limit is
2451reached.
2452
2453The number of tasks in a cgroup can be exhausted in ways which other
2454controllers cannot prevent, thus warranting its own controller.  For
2455example, a fork bomb is likely to exhaust the number of tasks before
2456hitting memory restrictions.
2457
2458Note that PIDs used in this controller refer to TIDs, process IDs as
2459used by the kernel.
2460
2461
2462PID Interface Files
2463~~~~~~~~~~~~~~~~~~~
2464
2465  pids.max
2466	A read-write single value file which exists on non-root
2467	cgroups.  The default is "max".
2468
2469	Hard limit of number of processes.
2470
2471  pids.current
2472	A read-only single value file which exists on non-root cgroups.
2473
2474	The number of processes currently in the cgroup and its
2475	descendants.
2476
2477  pids.peak
2478	A read-only single value file which exists on non-root cgroups.
2479
2480	The maximum value that the number of processes in the cgroup and its
2481	descendants has ever reached.
2482
2483  pids.events
2484	A read-only flat-keyed file which exists on non-root cgroups. Unless
2485	specified otherwise, a value change in this file generates a file
2486	modified event. The following entries are defined.
2487
2488	  max
2489		The number of times the cgroup's total number of processes hit the pids.max
2490		limit (see also pids_localevents).
2491
2492  pids.events.local
2493	Similar to pids.events but the fields in the file are local
2494	to the cgroup i.e. not hierarchical. The file modified event
2495	generated on this file reflects only the local events.
2496
2497Organisational operations are not blocked by cgroup policies, so it is
2498possible to have pids.current > pids.max.  This can be done by either
2499setting the limit to be smaller than pids.current, or attaching enough
2500processes to the cgroup such that pids.current is larger than
2501pids.max.  However, it is not possible to violate a cgroup PID policy
2502through fork() or clone(). These will return -EAGAIN if the creation
2503of a new process would cause a cgroup policy to be violated.
2504
2505
2506Cpuset
2507------
2508
2509The "cpuset" controller provides a mechanism for constraining
2510the CPU and memory node placement of tasks to only the resources
2511specified in the cpuset interface files in a task's current cgroup.
2512This is especially valuable on large NUMA systems where placing jobs
2513on properly sized subsets of the systems with careful processor and
2514memory placement to reduce cross-node memory access and contention
2515can improve overall system performance.
2516
2517The "cpuset" controller is hierarchical.  That means the controller
2518cannot use CPUs or memory nodes not allowed in its parent.
2519
2520
2521Cpuset Interface Files
2522~~~~~~~~~~~~~~~~~~~~~~
2523
2524  cpuset.cpus
2525	A read-write multiple values file which exists on non-root
2526	cpuset-enabled cgroups.
2527
2528	It lists the requested CPUs to be used by tasks within this
2529	cgroup.  The actual list of CPUs to be granted, however, is
2530	subjected to constraints imposed by its parent and can differ
2531	from the requested CPUs.
2532
2533	The CPU numbers are comma-separated numbers or ranges.
2534	For example::
2535
2536	  # cat cpuset.cpus
2537	  0-4,6,8-10
2538
2539	An empty value indicates that the cgroup is using the same
2540	setting as the nearest cgroup ancestor with a non-empty
2541	"cpuset.cpus" or all the available CPUs if none is found.
2542
2543	The value of "cpuset.cpus" stays constant until the next update
2544	and won't be affected by any CPU hotplug events.
2545
2546  cpuset.cpus.effective
2547	A read-only multiple values file which exists on all
2548	cpuset-enabled cgroups.
2549
2550	It lists the active CPUs available to tasks in this cgroup.
2551
2552	For a cgroup that is not a valid partition root, an empty
2553	"cpuset.cpus" makes "cpuset.cpus.effective" show all CPUs
2554	available from the parent cgroup. Otherwise, it is a subset
2555	of "cpuset.cpus" unless none of the requested CPUs can be
2556	granted. In that case, it is treated like an empty
2557	"cpuset.cpus".
2558
2559	For example::
2560
2561	  P (cpuset.cpus.effective=0-3)
2562	  |
2563	  +-- C0 (member):       cpuset.cpus=1-2
2564	  |                      cpuset.cpus.effective=1-2
2565	  +-- C1 (invalid root): cpuset.cpus=4-5
2566	  |                      cpuset.cpus.effective=0-3
2567	  \-- C2 (member):       cpuset.cpus=<empty>
2568	                         cpuset.cpus.effective=0-3
2569
2570	C0 gets CPUs 1-2 because they are available from P.
2571	None of C1's requested CPUs 4-5 are available from P, so C1 uses P's
2572	effective CPUs 0-3 instead.
2573	C2 has an empty "cpuset.cpus" and therefore also uses P's effective
2574	CPUs 0-3.
2575
2576	For a valid non-root partition root, "cpuset.cpus.effective" contains
2577	the active CPUs from "cpuset.cpus.exclusive.effective" except
2578	those granted to valid child partition roots. If
2579	"cpuset.cpus.exclusive" is set to a value different from
2580	"cpuset.cpus", the effective CPUs need not be a subset of
2581	"cpuset.cpus".
2582
2583	For example, if CPUs 2-5 are allocated to P for exclusive use,
2584	CPUs 2-4 are active, CPU 5 is offline, and none of CPUs 2-5 are
2585	assigned to a valid child partition root::
2586
2587	  top (root)
2588	  |
2589	  \-- P (valid partition root)
2590	      cpuset.cpus=0-3
2591	      cpuset.cpus.exclusive=2-5
2592	      cpuset.cpus.exclusive.effective=2-5
2593	      cpuset.cpus.effective=2-4
2594
2595	P has 2-5 in "cpuset.cpus.exclusive.effective", including the
2596	offline CPU 5. CPU 5 is excluded from "cpuset.cpus.effective",
2597	leaving 2-4. The result also shows that "cpuset.cpus.effective"
2598	need not be a subset of "cpuset.cpus", which is 0-3.
2599
2600	Its value will be affected by CPU hotplug events.
2601
2602  cpuset.mems
2603	A read-write multiple values file which exists on non-root
2604	cpuset-enabled cgroups.
2605
2606	It lists the requested memory nodes to be used by tasks within
2607	this cgroup.  The actual list of memory nodes granted, however,
2608	is subjected to constraints imposed by its parent and can differ
2609	from the requested memory nodes.
2610
2611	The memory node numbers are comma-separated numbers or ranges.
2612	For example::
2613
2614	  # cat cpuset.mems
2615	  0-1,3
2616
2617	An empty value indicates that the cgroup is using the same
2618	setting as the nearest cgroup ancestor with a non-empty
2619	"cpuset.mems" or all the available memory nodes if none
2620	is found.
2621
2622	The value of "cpuset.mems" stays constant until the next update
2623	and won't be affected by any memory nodes hotplug events.
2624
2625	Setting a non-empty value to "cpuset.mems" causes memory of
2626	tasks within the cgroup to be migrated to the designated nodes if
2627	they are currently using memory outside of the designated nodes.
2628
2629	There is a cost for this memory migration.  The migration
2630	may not be complete and some memory pages may be left behind.
2631	So it is recommended that "cpuset.mems" should be set properly
2632	before spawning new tasks into the cpuset.  Even if there is
2633	a need to change "cpuset.mems" with active tasks, it shouldn't
2634	be done frequently.
2635
2636	For a multithreaded process, the threadgroup leader is
2637	considered the owner of the group's memory. Memory policy
2638	rebinding and migration will only happen with respect to the
2639	threadgroup leader. To avoid unexpected results, non-leading
2640	threads shouldn't be put into another cgroup whose "cpuset.mems"
2641	doesn't fully overlap that of the threadgroup leader.
2642
2643  cpuset.mems.effective
2644	A read-only multiple values file which exists on all
2645	cpuset-enabled cgroups.
2646
2647	It lists the onlined memory nodes that are actually granted to
2648	this cgroup by its parent. These memory nodes are allowed to
2649	be used by tasks within the current cgroup.
2650
2651	If "cpuset.mems" is empty, it shows all the memory nodes from the
2652	parent cgroup that will be available to be used by this cgroup.
2653	Otherwise, it should be a subset of "cpuset.mems" unless none of
2654	the memory nodes listed in "cpuset.mems" can be granted.  In this
2655	case, it will be treated just like an empty "cpuset.mems".
2656
2657	Its value will be affected by memory nodes hotplug events.
2658
2659  cpuset.cpus.exclusive
2660	A read-write multiple values file which exists on non-root
2661	cpuset-enabled cgroups.
2662
2663	It lists all the exclusive CPUs that are allowed to be used
2664	to create a new cpuset partition.  Its value is not used
2665	unless the cgroup becomes a valid partition root.  See the
2666	"cpuset.cpus.partition" section below for a description of what
2667	a cpuset partition is.
2668
2669	When the cgroup becomes a partition root, the actual exclusive
2670	CPUs that are allocated to that partition are listed in
2671	"cpuset.cpus.exclusive.effective" which may be different
2672	from "cpuset.cpus.exclusive".  If "cpuset.cpus.exclusive"
2673	has previously been set, "cpuset.cpus.exclusive.effective"
2674	is always a subset of it.
2675
2676	Users can manually set it to a value that is different from
2677	"cpuset.cpus".	One constraint in setting it is that the list of
2678	CPUs must be exclusive with respect to "cpuset.cpus.exclusive"
2679	and "cpuset.cpus.exclusive.effective" of its siblings.	Another
2680	constraint is that it cannot be a superset of "cpuset.cpus"
2681	of its sibling in order to leave at least one CPU available to
2682	that sibling when the exclusive CPUs are taken away.
2683
2684	For a parent cgroup, any one of its exclusive CPUs can only
2685	be distributed to at most one of its child cgroups.  Having an
2686	exclusive CPU appearing in two or more of its child cgroups is
2687	not allowed (the exclusivity rule).  A value that violates the
2688	exclusivity rule will be rejected with a write error.
2689
2690	The root cgroup is a partition root and all its available CPUs
2691	are in its exclusive CPU set.
2692
2693  cpuset.cpus.exclusive.effective
2694	A read-only multiple values file which exists on all non-root
2695	cpuset-enabled cgroups.
2696
2697	This file shows the effective set of exclusive CPUs that
2698	can be used to create a partition root.  The content
2699	of this file will always be a subset of its parent's
2700	"cpuset.cpus.exclusive.effective" if its parent is not the root
2701	cgroup.  It will also be a subset of "cpuset.cpus.exclusive"
2702	if it is set.  This file should only be non-empty if either
2703	"cpuset.cpus.exclusive" is set or when the current cpuset is
2704	a valid partition root.
2705
2706  cpuset.cpus.isolated
2707	A read-only and root cgroup only multiple values file.
2708
2709	This file shows the set of all isolated CPUs used in existing
2710	isolated partitions or isolated at boot time via the "domain"
2711	flag of "isolcpus". The latter remain listed even after any
2712	partition using them is released.
2713
2714  cpuset.cpus.partition
2715	A read-write single value file which exists on non-root
2716	cpuset-enabled cgroups.  This flag is owned by the parent cgroup
2717	and is not delegatable.
2718
2719	It accepts only the following input values when written to.
2720
2721	  ==========	=====================================
2722	  "member"	Non-root member of a partition
2723	  "root"	Partition root
2724	  "isolated"	Partition root without load balancing
2725	  ==========	=====================================
2726
2727	A cpuset partition is a collection of cpuset-enabled cgroups with
2728	a partition root at the top of the hierarchy and its descendants
2729	except those that are separate partition roots themselves and
2730	their descendants.  A partition has exclusive access to the
2731	set of exclusive CPUs allocated to it.	Other cgroups outside
2732	of that partition cannot use any CPUs in that set.
2733
2734	There are two types of partitions - local and remote.  A local
2735	partition is one whose parent cgroup is also a valid partition
2736	root.  A remote partition is one whose parent cgroup is not a
2737	valid partition root itself.
2738
2739	Writing to "cpuset.cpus.exclusive" is optional for the creation
2740	of a local partition as its "cpuset.cpus.exclusive" file will
2741	assume an implicit value that is the same as "cpuset.cpus" if it
2742	is not set.  Writing the proper "cpuset.cpus.exclusive" values
2743	down the cgroup hierarchy before the target partition root is
2744	mandatory for the creation of a remote partition.
2745
2746	Not all the CPUs requested in "cpuset.cpus.exclusive" can be
2747	used to form a new partition.  Only those that were present
2748	in its parent's "cpuset.cpus.exclusive.effective" control
2749	file can be used.  For partitions created without setting
2750	"cpuset.cpus.exclusive", exclusive CPUs specified in sibling's
2751	"cpuset.cpus.exclusive" or "cpuset.cpus.exclusive.effective"
2752	also cannot be used.
2753
2754	Currently, a remote partition cannot be created under a local
2755	partition.  All the ancestors of a remote partition root except
2756	the root cgroup cannot be a partition root.
2757
2758	The root cgroup is always a partition root and its state cannot
2759	be changed.  All other non-root cgroups start out as "member".
2760	Even though the "cpuset.cpus.exclusive*" and "cpuset.cpus"
2761	control files are not present in the root cgroup, they are
2762	implicitly the same as the "/sys/devices/system/cpu/possible"
2763	sysfs file.
2764
2765	When set to "root", the current cgroup is the root of a new
2766	partition or scheduling domain.  The set of exclusive CPUs is
2767	determined by the value of its "cpuset.cpus.exclusive.effective".
2768
2769	When set to "isolated", the CPUs in that partition will be in
2770	an isolated state without any load balancing from the scheduler
2771	and excluded from the unbound workqueues.  Tasks placed in such
2772	a partition with multiple CPUs should be carefully distributed
2773	and bound to each of the individual CPUs for optimal performance.
2774
2775	A partition root ("root" or "isolated") can be in one of the
2776	two possible states - valid or invalid.  An invalid partition
2777	root is in a degraded state where some state information may
2778	be retained, but behaves more like a "member".
2779
2780	All possible state transitions among "member", "root" and
2781	"isolated" are allowed.
2782
2783	On read, the "cpuset.cpus.partition" file can show the following
2784	values.
2785
2786	  =============================	=====================================
2787	  "member"			Non-root member of a partition
2788	  "root"			Partition root
2789	  "isolated"			Partition root without load balancing
2790	  "root invalid (<reason>)"	Invalid partition root
2791	  "isolated invalid (<reason>)"	Invalid isolated partition root
2792	  =============================	=====================================
2793
2794	In the case of an invalid partition root, a descriptive string on
2795	why the partition is invalid is included within parentheses.
2796
2797	For a local partition root to be valid, the following conditions
2798	must be met.
2799
2800	1) The parent cgroup is a valid partition root.
2801	2) The "cpuset.cpus.exclusive.effective" file cannot be empty,
2802	   though it may contain offline CPUs.
2803	3) The "cpuset.cpus.effective" cannot be empty unless there is
2804	   no task associated with this partition.
2805
2806	For a remote partition root to be valid, all the above conditions
2807	except the first one must be met.
2808
2809	External events like hotplug or changes to "cpuset.cpus" or
2810	"cpuset.cpus.exclusive" can cause a valid partition root to
2811	become invalid and vice versa.	Note that a task cannot be
2812	moved to a cgroup with empty "cpuset.cpus.effective".
2813
2814	A valid non-root parent partition may distribute out all its CPUs
2815	to its child local partitions when there is no task associated
2816	with it.
2817
2818	Care must be taken to change a valid partition root to "member"
2819	as all its child local partitions, if present, will become
2820	invalid causing disruption to tasks running in those child
2821	partitions. These inactivated partitions could be recovered if
2822	their parent is switched back to a partition root with a proper
2823	value in "cpuset.cpus" or "cpuset.cpus.exclusive".
2824
2825	Poll and inotify events are triggered whenever the state of
2826	"cpuset.cpus.partition" changes.  That includes changes caused
2827	by write to "cpuset.cpus.partition", cpu hotplug or other
2828	changes that modify the validity status of the partition.
2829	This will allow user space agents to monitor unexpected changes
2830	to "cpuset.cpus.partition" without the need to do continuous
2831	polling.
2832
2833	A user can pre-configure certain CPUs to an isolated state
2834	with load balancing disabled at boot time with the "isolcpus"
2835	kernel boot command line option.  If those CPUs are to be put
2836	into a partition, they have to be used in an isolated partition.
2837
2838
2839Device controller
2840-----------------
2841
2842Device controller manages access to device files. It includes both
2843creation of new device files (using mknod), and access to the
2844existing device files.
2845
2846Cgroup v2 device controller has no interface files and is implemented
2847on top of cgroup BPF. To control access to device files, a user may
2848create bpf programs of type BPF_PROG_TYPE_CGROUP_DEVICE and attach
2849them to cgroups with BPF_CGROUP_DEVICE flag. On an attempt to access a
2850device file, corresponding BPF programs will be executed, and depending
2851on the return value the attempt will succeed or fail with -EPERM.
2852
2853A BPF_PROG_TYPE_CGROUP_DEVICE program takes a pointer to the
2854bpf_cgroup_dev_ctx structure, which describes the device access attempt:
2855access type (mknod/read/write) and device (type, major and minor numbers).
2856If the program returns 0, the attempt fails with -EPERM, otherwise it
2857succeeds.
2858
2859An example of BPF_PROG_TYPE_CGROUP_DEVICE program may be found in
2860tools/testing/selftests/bpf/progs/dev_cgroup.c in the kernel source tree.
2861
2862
2863RDMA
2864----
2865
2866The "rdma" controller regulates the distribution and accounting of
2867RDMA resources.
2868
2869RDMA devices from all network namespaces are listed. Each line starts with
2870the device name. If more than one device has the same name, ``index=N``
2871follows the name, where ``N`` is the system-wide RDMA device index, unique
2872among registered devices.
2873
2874RDMA Interface Files
2875~~~~~~~~~~~~~~~~~~~~
2876
2877  rdma.max
2878	A readwrite nested-keyed file that exists for all the cgroups
2879	except root that describes current configured resource limit
2880	for a RDMA/IB device.
2881
2882	Lines are keyed by device name and are not ordered. A write may
2883	include ``index=N`` after the device name. The index is optional
2884	when the name is globally unique. If multiple devices have that
2885	name, the index is required and a write without it fails with
2886	``-ENOTUNIQ``.
2887	Each line contains space separated resource name and its configured
2888	limit that can be distributed.
2889
2890	The following nested keys are defined.
2891
2892	  ==========	=============================
2893	  hca_handle	Maximum number of HCA Handles
2894	  hca_object 	Maximum number of HCA Objects
2895	  ==========	=============================
2896
2897	An example for mlx4 and ocrdma device follows::
2898
2899	  mlx4_0 hca_handle=2 hca_object=2000
2900	  ocrdma1 hca_handle=3 hca_object=max
2901
2902	For devices with duplicate names, select the device by index::
2903
2904	  echo "rxe0 index=5 hca_handle=2" > rdma.max
2905
2906  rdma.current
2907	A read-only file that describes current resource usage.
2908	It exists for all the cgroup except root.
2909
2910	An example for mlx4 and ocrdma device follows::
2911
2912	  mlx4_0 hca_handle=1 hca_object=20
2913	  ocrdma1 hca_handle=1 hca_object=23
2914
2915  rdma.peak
2916	A read-only nested-keyed file that exists for all the cgroups
2917	except root.  It shows the historical high watermark of
2918	resource usage per device since the cgroup was created.
2919
2920	An example for mlx4 and ocrdma device follows::
2921
2922	  mlx4_0 hca_handle=1 hca_object=20
2923	  ocrdma1 hca_handle=0 hca_object=23
2924
2925  rdma.events
2926	A read-only nested-keyed file which exists on non-root
2927	cgroups.  The following nested keys are defined.
2928
2929	  max
2930		The number of times a process in this cgroup or its
2931		descendants attempted an RDMA resource allocation that
2932		was rejected because a rdma.max limit in the subtree
2933		was reached.  This is a hierarchical counter: the event
2934		is propagated upward to all ancestor cgroups.  A value
2935		change in this file generates a file modified event.
2936
2937	  alloc_fail
2938		The number of RDMA resource allocation attempts that
2939		originated in this cgroup or its descendants and failed
2940		due to a rdma.max limit being reached.  This is a
2941		hierarchical counter propagated upward.
2942
2943	An example for mlx4 device follows::
2944
2945	  mlx4_0 hca_handle.max=5 hca_handle.alloc_fail=3 hca_object.max=0 hca_object.alloc_fail=0
2946
2947  rdma.events.local
2948	Similar to rdma.events but the fields in the file are local
2949	to the cgroup i.e. not hierarchical.  The file modified event
2950	generated on this file reflects only the local events.
2951
2952	The following nested keys are defined.
2953
2954	  max
2955		The number of times a process in this cgroup or its
2956		descendants attempted an RDMA resource allocation that
2957		was rejected because this cgroup's own rdma.max limit
2958		was reached.
2959	  alloc_fail
2960		The number of RDMA resource allocation attempts
2961		originating from this cgroup that failed due to this
2962		cgroup's or an ancestor's rdma.max limit.
2963
2964	An example for mlx4 device follows::
2965
2966	  mlx4_0 hca_handle.max=5 hca_handle.alloc_fail=0 hca_object.max=0 hca_object.alloc_fail=0
2967
2968DMEM
2969----
2970
2971The "dmem" controller regulates the distribution and accounting of
2972device memory regions. Because each memory region may have its own page size,
2973which does not have to be equal to the system page size, the units are always bytes.
2974
2975DMEM Interface Files
2976~~~~~~~~~~~~~~~~~~~~
2977
2978  dmem.max, dmem.min, dmem.low
2979	A readwrite nested-keyed file that exists for all the cgroups
2980	except root that describes current configured resource limit
2981	for a region.
2982
2983	An example for xe follows::
2984
2985	  drm/0000:03:00.0/vram0 1073741824
2986	  drm/0000:03:00.0/stolen max
2987
2988	The semantics are the same as for the memory cgroup controller, and are
2989	calculated in the same way.
2990
2991  dmem.peak
2992	A read-only nested-keyed file that exists on non-root cgroups.
2993
2994	The max device memory usage recorded for the cgroup and its
2995	descendants since the creation of the cgroup for each region.
2996
2997  dmem.capacity
2998	A read-only file that describes maximum region capacity.
2999	It only exists on the root cgroup. Not all memory can be
3000	allocated by cgroups, as the kernel reserves some for
3001	internal use.
3002
3003	An example for xe follows::
3004
3005	  drm/0000:03:00.0/vram0 8514437120
3006	  drm/0000:03:00.0/stolen 67108864
3007
3008  dmem.current
3009	A read-only file that describes current resource usage.
3010	It exists for all the cgroup except root.
3011
3012	An example for xe follows::
3013
3014	  drm/0000:03:00.0/vram0 12550144
3015	  drm/0000:03:00.0/stolen 8650752
3016
3017HugeTLB
3018-------
3019
3020The HugeTLB controller allows limiting the HugeTLB usage per control group and
3021enforces the controller limit during page fault.
3022
3023HugeTLB Interface Files
3024~~~~~~~~~~~~~~~~~~~~~~~
3025
3026  hugetlb.<hugepagesize>.current
3027	Show current usage for "hugepagesize" hugetlb.  It exists for all
3028	the cgroup except root.
3029
3030  hugetlb.<hugepagesize>.max
3031	Set/show the hard limit of "hugepagesize" hugetlb usage.
3032	The default value is "max".  It exists for all the cgroup except root.
3033
3034  hugetlb.<hugepagesize>.events
3035	A read-only flat-keyed file which exists on non-root cgroups.
3036
3037	  max
3038		The number of allocation failure due to HugeTLB limit
3039
3040  hugetlb.<hugepagesize>.events.local
3041	Similar to hugetlb.<hugepagesize>.events but the fields in the file
3042	are local to the cgroup i.e. not hierarchical. The file modified event
3043	generated on this file reflects only the local events.
3044
3045  hugetlb.<hugepagesize>.numa_stat
3046	Similar to memory.numa_stat, it shows the numa information of the
3047        hugetlb pages of <hugepagesize> in this cgroup.  Only active in
3048        use hugetlb pages are included.  The per-node values are in bytes.
3049
3050Misc
3051----
3052
3053The Miscellaneous cgroup provides the resource limiting and tracking
3054mechanism for the scalar resources which cannot be abstracted like the other
3055cgroup resources. Controller is enabled by the CONFIG_CGROUP_MISC config
3056option.
3057
3058A resource can be added to the controller via enum misc_res_type{} in the
3059include/linux/misc_cgroup.h file and the corresponding name via misc_res_name[]
3060in the kernel/cgroup/misc.c file. Provider of the resource must set its
3061capacity prior to using the resource by calling misc_cg_set_capacity().
3062
3063Once a capacity is set then the resource usage can be updated using charge and
3064uncharge APIs. All of the APIs to interact with misc controller are in
3065include/linux/misc_cgroup.h.
3066
3067Misc Interface Files
3068~~~~~~~~~~~~~~~~~~~~
3069
3070Miscellaneous controller provides the following interface files. If two misc
3071resources (res_a and res_b) are registered then:
3072
3073  misc.capacity
3074        A read-only flat-keyed file shown only in the root cgroup.  It shows
3075        miscellaneous scalar resources available on the platform along with
3076        their quantities::
3077
3078	  $ cat misc.capacity
3079	  res_a 50
3080	  res_b 10
3081
3082  misc.current
3083        A read-only flat-keyed file shown in the all cgroups.  It shows
3084        the current usage of the resources in the cgroup and its children.::
3085
3086	  $ cat misc.current
3087	  res_a 3
3088	  res_b 0
3089
3090  misc.peak
3091        A read-only flat-keyed file shown in all cgroups.  It shows the
3092        historical maximum usage of the resources in the cgroup and its
3093        children.::
3094
3095	  $ cat misc.peak
3096	  res_a 10
3097	  res_b 8
3098
3099  misc.max
3100        A read-write flat-keyed file shown in the non root cgroups. Allowed
3101        maximum usage of the resources in the cgroup and its children.::
3102
3103	  $ cat misc.max
3104	  res_a max
3105	  res_b 4
3106
3107	Limit can be set by::
3108
3109	  # echo res_a 1 > misc.max
3110
3111	Limit can be set to max by::
3112
3113	  # echo res_a max > misc.max
3114
3115        Limits can be set higher than the capacity value in the misc.capacity
3116        file.
3117
3118  misc.events
3119	A read-only flat-keyed file which exists on non-root cgroups. The
3120	following entries are defined. Unless specified otherwise, a value
3121	change in this file generates a file modified event. All fields in
3122	this file are hierarchical.
3123
3124	  <res>.max
3125		The number of times the cgroup's resource usage was
3126		about to go over the max boundary.
3127
3128  misc.events.local
3129        Similar to misc.events but the fields in the file are local to the
3130        cgroup i.e. not hierarchical. The file modified event generated on
3131        this file reflects only the local events.
3132
3133Migration and Ownership
3134~~~~~~~~~~~~~~~~~~~~~~~
3135
3136A miscellaneous scalar resource is charged to the cgroup in which it is used
3137first, and stays charged to that cgroup until that resource is freed. Migrating
3138a process to a different cgroup does not move the charge to the destination
3139cgroup where the process has moved.
3140
3141Others
3142------
3143
3144perf_event
3145~~~~~~~~~~
3146
3147perf_event controller, if not mounted on a legacy hierarchy, is
3148automatically enabled on the v2 hierarchy so that perf events can
3149always be filtered by cgroup v2 path.  The controller can still be
3150moved to a legacy hierarchy after v2 hierarchy is populated.
3151
3152
3153Non-normative information
3154-------------------------
3155
3156This section contains information that isn't considered to be a part of
3157the stable kernel API and so is subject to change.
3158
3159
3160CPU controller root cgroup process behaviour
3161~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
3162
3163When distributing CPU cycles in the root cgroup each thread in this
3164cgroup is treated as if it was hosted in a separate child cgroup of the
3165root cgroup. This child cgroup weight is dependent on its thread nice
3166level.
3167
3168For details of this mapping see sched_prio_to_weight array in
3169kernel/sched/core.c file (values from this array should be scaled
3170appropriately so the neutral - nice 0 - value is 100 instead of 1024).
3171
3172
3173IO controller root cgroup process behaviour
3174~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
3175
3176Root cgroup processes are hosted in an implicit leaf child node.
3177When distributing IO resources this implicit child node is taken into
3178account as if it was a normal child cgroup of the root cgroup with a
3179weight value of 200.
3180
3181
3182Namespace
3183=========
3184
3185Basics
3186------
3187
3188cgroup namespace provides a mechanism to virtualize the view of the
3189"/proc/$PID/cgroup" file and cgroup mounts.  The CLONE_NEWCGROUP clone
3190flag can be used with clone(2) and unshare(2) to create a new cgroup
3191namespace.  The process running inside the cgroup namespace will have
3192its "/proc/$PID/cgroup" output restricted to cgroupns root.  The
3193cgroupns root is the cgroup of the process at the time of creation of
3194the cgroup namespace.
3195
3196Without cgroup namespace, the "/proc/$PID/cgroup" file shows the
3197complete path of the cgroup of a process.  In a container setup where
3198a set of cgroups and namespaces are intended to isolate processes the
3199"/proc/$PID/cgroup" file may leak potential system level information
3200to the isolated processes.  For example::
3201
3202  # cat /proc/self/cgroup
3203  0::/batchjobs/container_id1
3204
3205The path '/batchjobs/container_id1' can be considered as system-data
3206and undesirable to expose to the isolated processes.  cgroup namespace
3207can be used to restrict visibility of this path.  For example, before
3208creating a cgroup namespace, one would see::
3209
3210  # ls -l /proc/self/ns/cgroup
3211  lrwxrwxrwx 1 root root 0 2014-07-15 10:37 /proc/self/ns/cgroup -> cgroup:[4026531835]
3212  # cat /proc/self/cgroup
3213  0::/batchjobs/container_id1
3214
3215After unsharing a new namespace, the view changes::
3216
3217  # ls -l /proc/self/ns/cgroup
3218  lrwxrwxrwx 1 root root 0 2014-07-15 10:35 /proc/self/ns/cgroup -> cgroup:[4026532183]
3219  # cat /proc/self/cgroup
3220  0::/
3221
3222When some thread from a multi-threaded process unshares its cgroup
3223namespace, the new cgroupns gets applied to the entire process (all
3224the threads).  This is natural for the v2 hierarchy; however, for the
3225legacy hierarchies, this may be unexpected.
3226
3227A cgroup namespace is alive as long as there are processes inside or
3228mounts pinning it.  When the last usage goes away, the cgroup
3229namespace is destroyed.  The cgroupns root and the actual cgroups
3230remain.
3231
3232
3233The Root and Views
3234------------------
3235
3236The 'cgroupns root' for a cgroup namespace is the cgroup in which the
3237process calling unshare(2) is running.  For example, if a process in
3238/batchjobs/container_id1 cgroup calls unshare, cgroup
3239/batchjobs/container_id1 becomes the cgroupns root.  For the
3240init_cgroup_ns, this is the real root ('/') cgroup.
3241
3242The cgroupns root cgroup does not change even if the namespace creator
3243process later moves to a different cgroup::
3244
3245  # ~/unshare -c # unshare cgroupns in some cgroup
3246  # cat /proc/self/cgroup
3247  0::/
3248  # mkdir sub_cgrp_1
3249  # echo 0 > sub_cgrp_1/cgroup.procs
3250  # cat /proc/self/cgroup
3251  0::/sub_cgrp_1
3252
3253Each process gets its namespace-specific view of "/proc/$PID/cgroup"
3254
3255Processes running inside the cgroup namespace will be able to see
3256cgroup paths (in /proc/self/cgroup) only inside their root cgroup.
3257From within an unshared cgroupns::
3258
3259  # sleep 100000 &
3260  [1] 7353
3261  # echo 7353 > sub_cgrp_1/cgroup.procs
3262  # cat /proc/7353/cgroup
3263  0::/sub_cgrp_1
3264
3265From the initial cgroup namespace, the real cgroup path will be
3266visible::
3267
3268  $ cat /proc/7353/cgroup
3269  0::/batchjobs/container_id1/sub_cgrp_1
3270
3271From a sibling cgroup namespace (that is, a namespace rooted at a
3272different cgroup), the cgroup path relative to its own cgroup
3273namespace root will be shown.  For instance, if PID 7353's cgroup
3274namespace root is at '/batchjobs/container_id2', then it will see::
3275
3276  # cat /proc/7353/cgroup
3277  0::/../container_id2/sub_cgrp_1
3278
3279Note that the relative path always starts with '/' to indicate that
3280its relative to the cgroup namespace root of the caller.
3281
3282
3283Migration and setns(2)
3284----------------------
3285
3286Processes inside a cgroup namespace can move into and out of the
3287namespace root if they have proper access to external cgroups.  For
3288example, from inside a namespace with cgroupns root at
3289/batchjobs/container_id1, and assuming that the global hierarchy is
3290still accessible inside cgroupns::
3291
3292  # cat /proc/7353/cgroup
3293  0::/sub_cgrp_1
3294  # echo 7353 > batchjobs/container_id2/cgroup.procs
3295  # cat /proc/7353/cgroup
3296  0::/../container_id2
3297
3298Note that this kind of setup is not encouraged.  A task inside cgroup
3299namespace should only be exposed to its own cgroupns hierarchy.
3300
3301setns(2) to another cgroup namespace is allowed when:
3302
3303(a) the process has CAP_SYS_ADMIN against its current user namespace
3304(b) the process has CAP_SYS_ADMIN against the target cgroup
3305    namespace's userns
3306
3307No implicit cgroup changes happen with attaching to another cgroup
3308namespace.  It is expected that the someone moves the attaching
3309process under the target cgroup namespace root.
3310
3311
3312Interaction with Other Namespaces
3313---------------------------------
3314
3315Namespace specific cgroup hierarchy can be mounted by a process
3316running inside a non-init cgroup namespace::
3317
3318  # mount -t cgroup2 none $MOUNT_POINT
3319
3320This will mount the unified cgroup hierarchy with cgroupns root as the
3321filesystem root.  The process needs CAP_SYS_ADMIN against its user and
3322mount namespaces.
3323
3324The virtualization of /proc/self/cgroup file combined with restricting
3325the view of cgroup hierarchy by namespace-private cgroupfs mount
3326provides a properly isolated cgroup view inside the container.
3327
3328
3329Information on Kernel Programming
3330=================================
3331
3332This section contains kernel programming information in the areas
3333where interacting with cgroup is necessary.  cgroup core and
3334controllers are not covered.
3335
3336
3337Filesystem Support for Writeback
3338--------------------------------
3339
3340A filesystem can support cgroup writeback by updating
3341address_space_operations->writepages() to annotate bio's using the
3342following two functions.
3343
3344  wbc_init_bio(@wbc, @bio)
3345	Should be called for each bio carrying writeback data and
3346	associates the bio with the inode's owner cgroup and the
3347	corresponding request queue.  This must be called after
3348	a queue (device) has been associated with the bio and
3349	before submission.
3350
3351  wbc_account_cgroup_owner(@wbc, @folio, @bytes)
3352	Should be called for each data segment being written out.
3353	While this function doesn't care exactly when it's called
3354	during the writeback session, it's the easiest and most
3355	natural to call it as data segments are added to a bio.
3356
3357With writeback bio's annotated, cgroup support can be enabled per
3358super_block by setting SB_I_CGROUPWB in ->s_iflags.  This allows for
3359selective disabling of cgroup writeback support which is helpful when
3360certain filesystem features, e.g. journaled data mode, are
3361incompatible.
3362
3363wbc_init_bio() binds the specified bio to its cgroup.  Depending on
3364the configuration, the bio may be executed at a lower priority and if
3365the writeback session is holding shared resources, e.g. a journal
3366entry, may lead to priority inversion.  There is no one easy solution
3367for the problem.  Filesystems can try to work around specific problem
3368cases by skipping wbc_init_bio() and using bio_associate_blkg()
3369directly.
3370
3371
3372Deprecated v1 Core Features
3373===========================
3374
3375- Multiple hierarchies including named ones are not supported.
3376
3377- All v1 mount options are not supported.
3378
3379- The "tasks" file is removed and "cgroup.procs" is not sorted.
3380
3381- "cgroup.clone_children" is removed.
3382
3383- /proc/cgroups is meaningless for v2.  Use "cgroup.controllers" or
3384  "cgroup.stat" files at the root instead.
3385
3386
3387Issues with v1 and Rationales for v2
3388====================================
3389
3390Multiple Hierarchies
3391--------------------
3392
3393cgroup v1 allowed an arbitrary number of hierarchies and each
3394hierarchy could host any number of controllers.  While this seemed to
3395provide a high level of flexibility, it wasn't useful in practice.
3396
3397For example, as there is only one instance of each controller, utility
3398type controllers such as freezer which can be useful in all
3399hierarchies could only be used in one.  The issue is exacerbated by
3400the fact that controllers couldn't be moved to another hierarchy once
3401hierarchies were populated.  Another issue was that all controllers
3402bound to a hierarchy were forced to have exactly the same view of the
3403hierarchy.  It wasn't possible to vary the granularity depending on
3404the specific controller.
3405
3406In practice, these issues heavily limited which controllers could be
3407put on the same hierarchy and most configurations resorted to putting
3408each controller on its own hierarchy.  Only closely related ones, such
3409as the cpu and cpuacct controllers, made sense to be put on the same
3410hierarchy.  This often meant that userland ended up managing multiple
3411similar hierarchies repeating the same steps on each hierarchy
3412whenever a hierarchy management operation was necessary.
3413
3414Furthermore, support for multiple hierarchies came at a steep cost.
3415It greatly complicated cgroup core implementation but more importantly
3416the support for multiple hierarchies restricted how cgroup could be
3417used in general and what controllers was able to do.
3418
3419There was no limit on how many hierarchies there might be, which meant
3420that a thread's cgroup membership couldn't be described in finite
3421length.  The key might contain any number of entries and was unlimited
3422in length, which made it highly awkward to manipulate and led to
3423addition of controllers which existed only to identify membership,
3424which in turn exacerbated the original problem of proliferating number
3425of hierarchies.
3426
3427Also, as a controller couldn't have any expectation regarding the
3428topologies of hierarchies other controllers might be on, each
3429controller had to assume that all other controllers were attached to
3430completely orthogonal hierarchies.  This made it impossible, or at
3431least very cumbersome, for controllers to cooperate with each other.
3432
3433In most use cases, putting controllers on hierarchies which are
3434completely orthogonal to each other isn't necessary.  What usually is
3435called for is the ability to have differing levels of granularity
3436depending on the specific controller.  In other words, hierarchy may
3437be collapsed from leaf towards root when viewed from specific
3438controllers.  For example, a given configuration might not care about
3439how memory is distributed beyond a certain level while still wanting
3440to control how CPU cycles are distributed.
3441
3442
3443Thread Granularity
3444------------------
3445
3446cgroup v1 allowed threads of a process to belong to different cgroups.
3447This didn't make sense for some controllers and those controllers
3448ended up implementing different ways to ignore such situations but
3449much more importantly it blurred the line between API exposed to
3450individual applications and system management interface.
3451
3452Generally, in-process knowledge is available only to the process
3453itself; thus, unlike service-level organization of processes,
3454categorizing threads of a process requires active participation from
3455the application which owns the target process.
3456
3457cgroup v1 had an ambiguously defined delegation model which got abused
3458in combination with thread granularity.  cgroups were delegated to
3459individual applications so that they can create and manage their own
3460sub-hierarchies and control resource distributions along them.  This
3461effectively raised cgroup to the status of a syscall-like API exposed
3462to lay programs.
3463
3464First of all, cgroup has a fundamentally inadequate interface to be
3465exposed this way.  For a process to access its own knobs, it has to
3466extract the path on the target hierarchy from /proc/self/cgroup,
3467construct the path by appending the name of the knob to the path, open
3468and then read and/or write to it.  This is not only extremely clunky
3469and unusual but also inherently racy.  There is no conventional way to
3470define transaction across the required steps and nothing can guarantee
3471that the process would actually be operating on its own sub-hierarchy.
3472
3473cgroup controllers implemented a number of knobs which would never be
3474accepted as public APIs because they were just adding control knobs to
3475system-management pseudo filesystem.  cgroup ended up with interface
3476knobs which were not properly abstracted or refined and directly
3477revealed kernel internal details.  These knobs got exposed to
3478individual applications through the ill-defined delegation mechanism
3479effectively abusing cgroup as a shortcut to implementing public APIs
3480without going through the required scrutiny.
3481
3482This was painful for both userland and kernel.  Userland ended up with
3483misbehaving and poorly abstracted interfaces and kernel exposing and
3484locked into constructs inadvertently.
3485
3486
3487Competition Between Inner Nodes and Threads
3488-------------------------------------------
3489
3490cgroup v1 allowed threads to be in any cgroups which created an
3491interesting problem where threads belonging to a parent cgroup and its
3492children cgroups competed for resources.  This was nasty as two
3493different types of entities competed and there was no obvious way to
3494settle it.  Different controllers did different things.
3495
3496The cpu controller considered threads and cgroups as equivalents and
3497mapped nice levels to cgroup weights.  This worked for some cases but
3498fell flat when children wanted to be allocated specific ratios of CPU
3499cycles and the number of internal threads fluctuated - the ratios
3500constantly changed as the number of competing entities fluctuated.
3501There also were other issues.  The mapping from nice level to weight
3502wasn't obvious or universal, and there were various other knobs which
3503simply weren't available for threads.
3504
3505The io controller implicitly created a hidden leaf node for each
3506cgroup to host the threads.  The hidden leaf had its own copies of all
3507the knobs with ``leaf_`` prefixed.  While this allowed equivalent
3508control over internal threads, it was with serious drawbacks.  It
3509always added an extra layer of nesting which wouldn't be necessary
3510otherwise, made the interface messy and significantly complicated the
3511implementation.
3512
3513The memory controller didn't have a way to control what happened
3514between internal tasks and child cgroups and the behavior was not
3515clearly defined.  There were attempts to add ad-hoc behaviors and
3516knobs to tailor the behavior to specific workloads which would have
3517led to problems extremely difficult to resolve in the long term.
3518
3519Multiple controllers struggled with internal tasks and came up with
3520different ways to deal with it; unfortunately, all the approaches were
3521severely flawed and, furthermore, the widely different behaviors
3522made cgroup as a whole highly inconsistent.
3523
3524This clearly is a problem which needs to be addressed from cgroup core
3525in a uniform way.
3526
3527
3528Other Interface Issues
3529----------------------
3530
3531cgroup v1 grew without oversight and developed a large number of
3532idiosyncrasies and inconsistencies.  One issue on the cgroup core side
3533was how an empty cgroup was notified - a userland helper binary was
3534forked and executed for each event.  The event delivery wasn't
3535recursive or delegatable.  The limitations of the mechanism also led
3536to in-kernel event delivery filtering mechanism further complicating
3537the interface.
3538
3539Controller interfaces were problematic too.  An extreme example is
3540controllers completely ignoring hierarchical organization and treating
3541all cgroups as if they were all located directly under the root
3542cgroup.  Some controllers exposed a large amount of inconsistent
3543implementation details to userland.
3544
3545There also was no consistency across controllers.  When a new cgroup
3546was created, some controllers defaulted to not imposing extra
3547restrictions while others disallowed any resource usage until
3548explicitly configured.  Configuration knobs for the same type of
3549control used widely differing naming schemes and formats.  Statistics
3550and information knobs were named arbitrarily and used different
3551formats and units even in the same controller.
3552
3553cgroup v2 establishes common conventions where appropriate and updates
3554controllers so that they expose minimal and consistent interfaces.
3555
3556
3557Controller Issues and Remedies
3558------------------------------
3559
3560Memory
3561~~~~~~
3562
3563The original lower boundary, the soft limit, is defined as a limit
3564that is per default unset.  As a result, the set of cgroups that
3565global reclaim prefers is opt-in, rather than opt-out.  The costs for
3566optimizing these mostly negative lookups are so high that the
3567implementation, despite its enormous size, does not even provide the
3568basic desirable behavior.  First off, the soft limit has no
3569hierarchical meaning.  All configured groups are organized in a global
3570rbtree and treated like equal peers, regardless where they are located
3571in the hierarchy.  This makes subtree delegation impossible.  Second,
3572the soft limit reclaim pass is so aggressive that it not just
3573introduces high allocation latencies into the system, but also impacts
3574system performance due to overreclaim, to the point where the feature
3575becomes self-defeating.
3576
3577The memory.low boundary on the other hand is a top-down allocated
3578reserve.  A cgroup enjoys reclaim protection when it's within its
3579effective low, which makes delegation of subtrees possible. It also
3580enjoys having reclaim pressure proportional to its overage when
3581above its effective low.
3582
3583The original high boundary, the hard limit, is defined as a strict
3584limit that can not budge, even if the OOM killer has to be called.
3585But this generally goes against the goal of making the most out of the
3586available memory.  The memory consumption of workloads varies during
3587runtime, and that requires users to overcommit.  But doing that with a
3588strict upper limit requires either a fairly accurate prediction of the
3589working set size or adding slack to the limit.  Since working set size
3590estimation is hard and error prone, and getting it wrong results in
3591OOM kills, most users tend to err on the side of a looser limit and
3592end up wasting precious resources.
3593
3594The memory.high boundary on the other hand can be set much more
3595conservatively.  When hit, it throttles allocations by forcing them
3596into direct reclaim to work off the excess, but it never invokes the
3597OOM killer.  As a result, a high boundary that is chosen too
3598aggressively will not terminate the processes, but instead it will
3599lead to gradual performance degradation.  The user can monitor this
3600and make corrections until the minimal memory footprint that still
3601gives acceptable performance is found.
3602
3603In extreme cases, with many concurrent allocations and a complete
3604breakdown of reclaim progress within the group, the high boundary can
3605be exceeded.  But even then it's mostly better to satisfy the
3606allocation from the slack available in other groups or the rest of the
3607system than killing the group.  Otherwise, memory.max is there to
3608limit this type of spillover and ultimately contain buggy or even
3609malicious applications.
3610
3611Setting the original memory.limit_in_bytes below the current usage was
3612subject to a race condition, where concurrent charges could cause the
3613limit setting to fail. memory.max on the other hand will first set the
3614limit to prevent new charges, and then reclaim and OOM kill until the
3615new limit is met - or the task writing to memory.max is killed.
3616
3617The combined memory+swap accounting and limiting is replaced by real
3618control over swap space.
3619
3620The main argument for a combined memory+swap facility in the original
3621cgroup design was that global or parental pressure would always be
3622able to swap all anonymous memory of a child group, regardless of the
3623child's own (possibly untrusted) configuration.  However, untrusted
3624groups can sabotage swapping by other means - such as referencing its
3625anonymous memory in a tight loop - and an admin can not assume full
3626swappability when overcommitting untrusted jobs.
3627
3628For trusted jobs, on the other hand, a combined counter is not an
3629intuitive userspace interface, and it flies in the face of the idea
3630that cgroup controllers should account and limit specific physical
3631resources.  Swap space is a resource like all others in the system,
3632and that's why unified hierarchy allows distributing it separately.
3633