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