Based on kernel version 2.6.27. Page generated on 2008-10-13 09:53 EST.
1 CGROUPS 2 ------- 3 4 Written by Paul Menage <menage[AT]google.com> based on Documentation/cpusets[DOT]txt 5 6 Original copyright statements from cpusets.txt: 7 Portions Copyright (C) 2004 BULL SA. 8 Portions Copyright (c) 2004-2006 Silicon Graphics, Inc. 9 Modified by Paul Jackson <pj[AT]sgi[DOT]com> 10 Modified by Christoph Lameter <clameter[AT]sgi[DOT]com> 11 12 CONTENTS: 13 ========= 14 15 1. Control Groups 16 1.1 What are cgroups ? 17 1.2 Why are cgroups needed ? 18 1.3 How are cgroups implemented ? 19 1.4 What does notify_on_release do ? 20 1.5 How do I use cgroups ? 21 2. Usage Examples and Syntax 22 2.1 Basic Usage 23 2.2 Attaching processes 24 3. Kernel API 25 3.1 Overview 26 3.2 Synchronization 27 3.3 Subsystem API 28 4. Questions 29 30 1. Control Groups 31 ================= 32 33 1.1 What are cgroups ? 34 ---------------------- 35 36 Control Groups provide a mechanism for aggregating/partitioning sets of 37 tasks, and all their future children, into hierarchical groups with 38 specialized behaviour. 39 40 Definitions: 41 42 A *cgroup* associates a set of tasks with a set of parameters for one 43 or more subsystems. 44 45 A *subsystem* is a module that makes use of the task grouping 46 facilities provided by cgroups to treat groups of tasks in 47 particular ways. A subsystem is typically a "resource controller" that 48 schedules a resource or applies per-cgroup limits, but it may be 49 anything that wants to act on a group of processes, e.g. a 50 virtualization subsystem. 51 52 A *hierarchy* is a set of cgroups arranged in a tree, such that 53 every task in the system is in exactly one of the cgroups in the 54 hierarchy, and a set of subsystems; each subsystem has system-specific 55 state attached to each cgroup in the hierarchy. Each hierarchy has 56 an instance of the cgroup virtual filesystem associated with it. 57 58 At any one time there may be multiple active hierachies of task 59 cgroups. Each hierarchy is a partition of all tasks in the system. 60 61 User level code may create and destroy cgroups by name in an 62 instance of the cgroup virtual file system, specify and query to 63 which cgroup a task is assigned, and list the task pids assigned to 64 a cgroup. Those creations and assignments only affect the hierarchy 65 associated with that instance of the cgroup file system. 66 67 On their own, the only use for cgroups is for simple job 68 tracking. The intention is that other subsystems hook into the generic 69 cgroup support to provide new attributes for cgroups, such as 70 accounting/limiting the resources which processes in a cgroup can 71 access. For example, cpusets (see Documentation/cpusets.txt) allows 72 you to associate a set of CPUs and a set of memory nodes with the 73 tasks in each cgroup. 74 75 1.2 Why are cgroups needed ? 76 ---------------------------- 77 78 There are multiple efforts to provide process aggregations in the 79 Linux kernel, mainly for resource tracking purposes. Such efforts 80 include cpusets, CKRM/ResGroups, UserBeanCounters, and virtual server 81 namespaces. These all require the basic notion of a 82 grouping/partitioning of processes, with newly forked processes ending 83 in the same group (cgroup) as their parent process. 84 85 The kernel cgroup patch provides the minimum essential kernel 86 mechanisms required to efficiently implement such groups. It has 87 minimal impact on the system fast paths, and provides hooks for 88 specific subsystems such as cpusets to provide additional behaviour as 89 desired. 90 91 Multiple hierarchy support is provided to allow for situations where 92 the division of tasks into cgroups is distinctly different for 93 different subsystems - having parallel hierarchies allows each 94 hierarchy to be a natural division of tasks, without having to handle 95 complex combinations of tasks that would be present if several 96 unrelated subsystems needed to be forced into the same tree of 97 cgroups. 98 99 At one extreme, each resource controller or subsystem could be in a 100 separate hierarchy; at the other extreme, all subsystems 101 would be attached to the same hierarchy. 102 103 As an example of a scenario (originally proposed by vatsa[AT]in.ibm[DOT]com) 104 that can benefit from multiple hierarchies, consider a large 105 university server with various users - students, professors, system 106 tasks etc. The resource planning for this server could be along the 107 following lines: 108 109 CPU : Top cpuset 110 / \ 111 CPUSet1 CPUSet2 112 | | 113 (Profs) (Students) 114 115 In addition (system tasks) are attached to topcpuset (so 116 that they can run anywhere) with a limit of 20% 117 118 Memory : Professors (50%), students (30%), system (20%) 119 120 Disk : Prof (50%), students (30%), system (20%) 121 122 Network : WWW browsing (20%), Network File System (60%), others (20%) 123 / \ 124 Prof (15%) students (5%) 125 126 Browsers like firefox/lynx go into the WWW network class, while (k)nfsd go 127 into NFS network class. 128 129 At the same time firefox/lynx will share an appropriate CPU/Memory class 130 depending on who launched it (prof/student). 131 132 With the ability to classify tasks differently for different resources 133 (by putting those resource subsystems in different hierarchies) then 134 the admin can easily set up a script which receives exec notifications 135 and depending on who is launching the browser he can 136 137 # echo browser_pid > /mnt/<restype>/<userclass>/tasks 138 139 With only a single hierarchy, he now would potentially have to create 140 a separate cgroup for every browser launched and associate it with 141 approp network and other resource class. This may lead to 142 proliferation of such cgroups. 143 144 Also lets say that the administrator would like to give enhanced network 145 access temporarily to a student's browser (since it is night and the user 146 wants to do online gaming :)) OR give one of the students simulation 147 apps enhanced CPU power, 148 149 With ability to write pids directly to resource classes, it's just a 150 matter of : 151 152 # echo pid > /mnt/network/<new_class>/tasks 153 (after some time) 154 # echo pid > /mnt/network/<orig_class>/tasks 155 156 Without this ability, he would have to split the cgroup into 157 multiple separate ones and then associate the new cgroups with the 158 new resource classes. 159 160 161 162 1.3 How are cgroups implemented ? 163 --------------------------------- 164 165 Control Groups extends the kernel as follows: 166 167 - Each task in the system has a reference-counted pointer to a 168 css_set. 169 170 - A css_set contains a set of reference-counted pointers to 171 cgroup_subsys_state objects, one for each cgroup subsystem 172 registered in the system. There is no direct link from a task to 173 the cgroup of which it's a member in each hierarchy, but this 174 can be determined by following pointers through the 175 cgroup_subsys_state objects. This is because accessing the 176 subsystem state is something that's expected to happen frequently 177 and in performance-critical code, whereas operations that require a 178 task's actual cgroup assignments (in particular, moving between 179 cgroups) are less common. A linked list runs through the cg_list 180 field of each task_struct using the css_set, anchored at 181 css_set->tasks. 182 183 - A cgroup hierarchy filesystem can be mounted for browsing and 184 manipulation from user space. 185 186 - You can list all the tasks (by pid) attached to any cgroup. 187 188 The implementation of cgroups requires a few, simple hooks 189 into the rest of the kernel, none in performance critical paths: 190 191 - in init/main.c, to initialize the root cgroups and initial 192 css_set at system boot. 193 194 - in fork and exit, to attach and detach a task from its css_set. 195 196 In addition a new file system, of type "cgroup" may be mounted, to 197 enable browsing and modifying the cgroups presently known to the 198 kernel. When mounting a cgroup hierarchy, you may specify a 199 comma-separated list of subsystems to mount as the filesystem mount 200 options. By default, mounting the cgroup filesystem attempts to 201 mount a hierarchy containing all registered subsystems. 202 203 If an active hierarchy with exactly the same set of subsystems already 204 exists, it will be reused for the new mount. If no existing hierarchy 205 matches, and any of the requested subsystems are in use in an existing 206 hierarchy, the mount will fail with -EBUSY. Otherwise, a new hierarchy 207 is activated, associated with the requested subsystems. 208 209 It's not currently possible to bind a new subsystem to an active 210 cgroup hierarchy, or to unbind a subsystem from an active cgroup 211 hierarchy. This may be possible in future, but is fraught with nasty 212 error-recovery issues. 213 214 When a cgroup filesystem is unmounted, if there are any 215 child cgroups created below the top-level cgroup, that hierarchy 216 will remain active even though unmounted; if there are no 217 child cgroups then the hierarchy will be deactivated. 218 219 No new system calls are added for cgroups - all support for 220 querying and modifying cgroups is via this cgroup file system. 221 222 Each task under /proc has an added file named 'cgroup' displaying, 223 for each active hierarchy, the subsystem names and the cgroup name 224 as the path relative to the root of the cgroup file system. 225 226 Each cgroup is represented by a directory in the cgroup file system 227 containing the following files describing that cgroup: 228 229 - tasks: list of tasks (by pid) attached to that cgroup 230 - releasable flag: cgroup currently removeable? 231 - notify_on_release flag: run the release agent on exit? 232 - release_agent: the path to use for release notifications (this file 233 exists in the top cgroup only) 234 235 Other subsystems such as cpusets may add additional files in each 236 cgroup dir. 237 238 New cgroups are created using the mkdir system call or shell 239 command. The properties of a cgroup, such as its flags, are 240 modified by writing to the appropriate file in that cgroups 241 directory, as listed above. 242 243 The named hierarchical structure of nested cgroups allows partitioning 244 a large system into nested, dynamically changeable, "soft-partitions". 245 246 The attachment of each task, automatically inherited at fork by any 247 children of that task, to a cgroup allows organizing the work load 248 on a system into related sets of tasks. A task may be re-attached to 249 any other cgroup, if allowed by the permissions on the necessary 250 cgroup file system directories. 251 252 When a task is moved from one cgroup to another, it gets a new 253 css_set pointer - if there's an already existing css_set with the 254 desired collection of cgroups then that group is reused, else a new 255 css_set is allocated. Note that the current implementation uses a 256 linear search to locate an appropriate existing css_set, so isn't 257 very efficient. A future version will use a hash table for better 258 performance. 259 260 To allow access from a cgroup to the css_sets (and hence tasks) 261 that comprise it, a set of cg_cgroup_link objects form a lattice; 262 each cg_cgroup_link is linked into a list of cg_cgroup_links for 263 a single cgroup on its cgrp_link_list field, and a list of 264 cg_cgroup_links for a single css_set on its cg_link_list. 265 266 Thus the set of tasks in a cgroup can be listed by iterating over 267 each css_set that references the cgroup, and sub-iterating over 268 each css_set's task set. 269 270 The use of a Linux virtual file system (vfs) to represent the 271 cgroup hierarchy provides for a familiar permission and name space 272 for cgroups, with a minimum of additional kernel code. 273 274 1.4 What does notify_on_release do ? 275 ------------------------------------ 276 277 If the notify_on_release flag is enabled (1) in a cgroup, then 278 whenever the last task in the cgroup leaves (exits or attaches to 279 some other cgroup) and the last child cgroup of that cgroup 280 is removed, then the kernel runs the command specified by the contents 281 of the "release_agent" file in that hierarchy's root directory, 282 supplying the pathname (relative to the mount point of the cgroup 283 file system) of the abandoned cgroup. This enables automatic 284 removal of abandoned cgroups. The default value of 285 notify_on_release in the root cgroup at system boot is disabled 286 (0). The default value of other cgroups at creation is the current 287 value of their parents notify_on_release setting. The default value of 288 a cgroup hierarchy's release_agent path is empty. 289 290 1.5 How do I use cgroups ? 291 -------------------------- 292 293 To start a new job that is to be contained within a cgroup, using 294 the "cpuset" cgroup subsystem, the steps are something like: 295 296 1) mkdir /dev/cgroup 297 2) mount -t cgroup -ocpuset cpuset /dev/cgroup 298 3) Create the new cgroup by doing mkdir's and write's (or echo's) in 299 the /dev/cgroup virtual file system. 300 4) Start a task that will be the "founding father" of the new job. 301 5) Attach that task to the new cgroup by writing its pid to the 302 /dev/cgroup tasks file for that cgroup. 303 6) fork, exec or clone the job tasks from this founding father task. 304 305 For example, the following sequence of commands will setup a cgroup 306 named "Charlie", containing just CPUs 2 and 3, and Memory Node 1, 307 and then start a subshell 'sh' in that cgroup: 308 309 mount -t cgroup cpuset -ocpuset /dev/cgroup 310 cd /dev/cgroup 311 mkdir Charlie 312 cd Charlie 313 /bin/echo 2-3 > cpuset.cpus 314 /bin/echo 1 > cpuset.mems 315 /bin/echo $$ > tasks 316 sh 317 # The subshell 'sh' is now running in cgroup Charlie 318 # The next line should display '/Charlie' 319 cat /proc/self/cgroup 320 321 2. Usage Examples and Syntax 322 ============================ 323 324 2.1 Basic Usage 325 --------------- 326 327 Creating, modifying, using the cgroups can be done through the cgroup 328 virtual filesystem. 329 330 To mount a cgroup hierarchy will all available subsystems, type: 331 # mount -t cgroup xxx /dev/cgroup 332 333 The "xxx" is not interpreted by the cgroup code, but will appear in 334 /proc/mounts so may be any useful identifying string that you like. 335 336 To mount a cgroup hierarchy with just the cpuset and numtasks 337 subsystems, type: 338 # mount -t cgroup -o cpuset,numtasks hier1 /dev/cgroup 339 340 To change the set of subsystems bound to a mounted hierarchy, just 341 remount with different options: 342 343 # mount -o remount,cpuset,ns /dev/cgroup 344 345 Note that changing the set of subsystems is currently only supported 346 when the hierarchy consists of a single (root) cgroup. Supporting 347 the ability to arbitrarily bind/unbind subsystems from an existing 348 cgroup hierarchy is intended to be implemented in the future. 349 350 Then under /dev/cgroup you can find a tree that corresponds to the 351 tree of the cgroups in the system. For instance, /dev/cgroup 352 is the cgroup that holds the whole system. 353 354 If you want to create a new cgroup under /dev/cgroup: 355 # cd /dev/cgroup 356 # mkdir my_cgroup 357 358 Now you want to do something with this cgroup. 359 # cd my_cgroup 360 361 In this directory you can find several files: 362 # ls 363 notify_on_release releasable tasks 364 (plus whatever files added by the attached subsystems) 365 366 Now attach your shell to this cgroup: 367 # /bin/echo $$ > tasks 368 369 You can also create cgroups inside your cgroup by using mkdir in this 370 directory. 371 # mkdir my_sub_cs 372 373 To remove a cgroup, just use rmdir: 374 # rmdir my_sub_cs 375 376 This will fail if the cgroup is in use (has cgroups inside, or 377 has processes attached, or is held alive by other subsystem-specific 378 reference). 379 380 2.2 Attaching processes 381 ----------------------- 382 383 # /bin/echo PID > tasks 384 385 Note that it is PID, not PIDs. You can only attach ONE task at a time. 386 If you have several tasks to attach, you have to do it one after another: 387 388 # /bin/echo PID1 > tasks 389 # /bin/echo PID2 > tasks 390 ... 391 # /bin/echo PIDn > tasks 392 393 You can attach the current shell task by echoing 0: 394 395 # echo 0 > tasks 396 397 3. Kernel API 398 ============= 399 400 3.1 Overview 401 ------------ 402 403 Each kernel subsystem that wants to hook into the generic cgroup 404 system needs to create a cgroup_subsys object. This contains 405 various methods, which are callbacks from the cgroup system, along 406 with a subsystem id which will be assigned by the cgroup system. 407 408 Other fields in the cgroup_subsys object include: 409 410 - subsys_id: a unique array index for the subsystem, indicating which 411 entry in cgroup->subsys this subsystem should be managing. 412 413 - name: should be initialized to a unique subsystem name. Should be 414 no longer than MAX_CGROUP_TYPE_NAMELEN. 415 416 - early_init: indicate if the subsystem needs early initialization 417 at system boot. 418 419 Each cgroup object created by the system has an array of pointers, 420 indexed by subsystem id; this pointer is entirely managed by the 421 subsystem; the generic cgroup code will never touch this pointer. 422 423 3.2 Synchronization 424 ------------------- 425 426 There is a global mutex, cgroup_mutex, used by the cgroup 427 system. This should be taken by anything that wants to modify a 428 cgroup. It may also be taken to prevent cgroups from being 429 modified, but more specific locks may be more appropriate in that 430 situation. 431 432 See kernel/cgroup.c for more details. 433 434 Subsystems can take/release the cgroup_mutex via the functions 435 cgroup_lock()/cgroup_unlock(). 436 437 Accessing a task's cgroup pointer may be done in the following ways: 438 - while holding cgroup_mutex 439 - while holding the task's alloc_lock (via task_lock()) 440 - inside an rcu_read_lock() section via rcu_dereference() 441 442 3.3 Subsystem API 443 ----------------- 444 445 Each subsystem should: 446 447 - add an entry in linux/cgroup_subsys.h 448 - define a cgroup_subsys object called <name>_subsys 449 450 Each subsystem may export the following methods. The only mandatory 451 methods are create/destroy. Any others that are null are presumed to 452 be successful no-ops. 453 454 struct cgroup_subsys_state *create(struct cgroup_subsys *ss, 455 struct cgroup *cgrp) 456 (cgroup_mutex held by caller) 457 458 Called to create a subsystem state object for a cgroup. The 459 subsystem should allocate its subsystem state object for the passed 460 cgroup, returning a pointer to the new object on success or a 461 negative error code. On success, the subsystem pointer should point to 462 a structure of type cgroup_subsys_state (typically embedded in a 463 larger subsystem-specific object), which will be initialized by the 464 cgroup system. Note that this will be called at initialization to 465 create the root subsystem state for this subsystem; this case can be 466 identified by the passed cgroup object having a NULL parent (since 467 it's the root of the hierarchy) and may be an appropriate place for 468 initialization code. 469 470 void destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) 471 (cgroup_mutex held by caller) 472 473 The cgroup system is about to destroy the passed cgroup; the subsystem 474 should do any necessary cleanup and free its subsystem state 475 object. By the time this method is called, the cgroup has already been 476 unlinked from the file system and from the child list of its parent; 477 cgroup->parent is still valid. (Note - can also be called for a 478 newly-created cgroup if an error occurs after this subsystem's 479 create() method has been called for the new cgroup). 480 481 void pre_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp); 482 (cgroup_mutex held by caller) 483 484 Called before checking the reference count on each subsystem. This may 485 be useful for subsystems which have some extra references even if 486 there are not tasks in the cgroup. 487 488 int can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, 489 struct task_struct *task) 490 (cgroup_mutex held by caller) 491 492 Called prior to moving a task into a cgroup; if the subsystem 493 returns an error, this will abort the attach operation. If a NULL 494 task is passed, then a successful result indicates that *any* 495 unspecified task can be moved into the cgroup. Note that this isn't 496 called on a fork. If this method returns 0 (success) then this should 497 remain valid while the caller holds cgroup_mutex. 498 499 void attach(struct cgroup_subsys *ss, struct cgroup *cgrp, 500 struct cgroup *old_cgrp, struct task_struct *task) 501 502 Called after the task has been attached to the cgroup, to allow any 503 post-attachment activity that requires memory allocations or blocking. 504 505 void fork(struct cgroup_subsy *ss, struct task_struct *task) 506 507 Called when a task is forked into a cgroup. 508 509 void exit(struct cgroup_subsys *ss, struct task_struct *task) 510 511 Called during task exit. 512 513 int populate(struct cgroup_subsys *ss, struct cgroup *cgrp) 514 515 Called after creation of a cgroup to allow a subsystem to populate 516 the cgroup directory with file entries. The subsystem should make 517 calls to cgroup_add_file() with objects of type cftype (see 518 include/linux/cgroup.h for details). Note that although this 519 method can return an error code, the error code is currently not 520 always handled well. 521 522 void post_clone(struct cgroup_subsys *ss, struct cgroup *cgrp) 523 524 Called at the end of cgroup_clone() to do any paramater 525 initialization which might be required before a task could attach. For 526 example in cpusets, no task may attach before 'cpus' and 'mems' are set 527 up. 528 529 void bind(struct cgroup_subsys *ss, struct cgroup *root) 530 (cgroup_mutex held by caller) 531 532 Called when a cgroup subsystem is rebound to a different hierarchy 533 and root cgroup. Currently this will only involve movement between 534 the default hierarchy (which never has sub-cgroups) and a hierarchy 535 that is being created/destroyed (and hence has no sub-cgroups). 536 537 4. Questions 538 ============ 539 540 Q: what's up with this '/bin/echo' ? 541 A: bash's builtin 'echo' command does not check calls to write() against 542 errors. If you use it in the cgroup file system, you won't be 543 able to tell whether a command succeeded or failed. 544 545 Q: When I attach processes, only the first of the line gets really attached ! 546 A: We can only return one error code per call to write(). So you should also 547 put only ONE pid.