]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - include/linux/cgroup-defs.h
c68ae7f0fb5f3900967441406a845fc451df685d
[mirror_ubuntu-artful-kernel.git] / include / linux / cgroup-defs.h
1 /*
2 * linux/cgroup-defs.h - basic definitions for cgroup
3 *
4 * This file provides basic type and interface. Include this file directly
5 * only if necessary to avoid cyclic dependencies.
6 */
7 #ifndef _LINUX_CGROUP_DEFS_H
8 #define _LINUX_CGROUP_DEFS_H
9
10 #include <linux/limits.h>
11 #include <linux/list.h>
12 #include <linux/idr.h>
13 #include <linux/wait.h>
14 #include <linux/mutex.h>
15 #include <linux/rcupdate.h>
16 #include <linux/percpu-refcount.h>
17 #include <linux/percpu-rwsem.h>
18 #include <linux/workqueue.h>
19
20 #ifdef CONFIG_CGROUPS
21
22 struct cgroup;
23 struct cgroup_root;
24 struct cgroup_subsys;
25 struct cgroup_taskset;
26 struct kernfs_node;
27 struct kernfs_ops;
28 struct kernfs_open_file;
29 struct seq_file;
30
31 #define MAX_CGROUP_TYPE_NAMELEN 32
32 #define MAX_CGROUP_ROOT_NAMELEN 64
33 #define MAX_CFTYPE_NAME 64
34
35 /* define the enumeration of all cgroup subsystems */
36 #define SUBSYS(_x) _x ## _cgrp_id,
37 enum cgroup_subsys_id {
38 #include <linux/cgroup_subsys.h>
39 CGROUP_SUBSYS_COUNT,
40 };
41 #undef SUBSYS
42
43 /* bits in struct cgroup_subsys_state flags field */
44 enum {
45 CSS_NO_REF = (1 << 0), /* no reference counting for this css */
46 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
47 CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */
48 };
49
50 /* bits in struct cgroup flags field */
51 enum {
52 /* Control Group requires release notifications to userspace */
53 CGRP_NOTIFY_ON_RELEASE,
54 /*
55 * Clone the parent's configuration when creating a new child
56 * cpuset cgroup. For historical reasons, this option can be
57 * specified at mount time and thus is implemented here.
58 */
59 CGRP_CPUSET_CLONE_CHILDREN,
60 };
61
62 /* cgroup_root->flags */
63 enum {
64 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
65 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
66 };
67
68 /* cftype->flags */
69 enum {
70 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
71 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
72 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
73 CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */
74
75 /* internal flags, do not use outside cgroup core proper */
76 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */
77 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */
78 };
79
80 /*
81 * cgroup_file is the handle for a file instance created in a cgroup which
82 * is used, for example, to generate file changed notifications. This can
83 * be obtained by setting cftype->file_offset.
84 */
85 struct cgroup_file {
86 /* do not access any fields from outside cgroup core */
87 struct kernfs_node *kn;
88 };
89
90 /*
91 * Per-subsystem/per-cgroup state maintained by the system. This is the
92 * fundamental structural building block that controllers deal with.
93 *
94 * Fields marked with "PI:" are public and immutable and may be accessed
95 * directly without synchronization.
96 */
97 struct cgroup_subsys_state {
98 /* PI: the cgroup that this css is attached to */
99 struct cgroup *cgroup;
100
101 /* PI: the cgroup subsystem that this css is attached to */
102 struct cgroup_subsys *ss;
103
104 /* reference count - access via css_[try]get() and css_put() */
105 struct percpu_ref refcnt;
106
107 /* PI: the parent css */
108 struct cgroup_subsys_state *parent;
109
110 /* siblings list anchored at the parent's ->children */
111 struct list_head sibling;
112 struct list_head children;
113
114 /*
115 * PI: Subsys-unique ID. 0 is unused and root is always 1. The
116 * matching css can be looked up using css_from_id().
117 */
118 int id;
119
120 unsigned int flags;
121
122 /*
123 * Monotonically increasing unique serial number which defines a
124 * uniform order among all csses. It's guaranteed that all
125 * ->children lists are in the ascending order of ->serial_nr and
126 * used to allow interrupting and resuming iterations.
127 */
128 u64 serial_nr;
129
130 /*
131 * Incremented by online self and children. Used to guarantee that
132 * parents are not offlined before their children.
133 */
134 atomic_t online_cnt;
135
136 /* percpu_ref killing and RCU release */
137 struct rcu_head rcu_head;
138 struct work_struct destroy_work;
139 };
140
141 /*
142 * A css_set is a structure holding pointers to a set of
143 * cgroup_subsys_state objects. This saves space in the task struct
144 * object and speeds up fork()/exit(), since a single inc/dec and a
145 * list_add()/del() can bump the reference count on the entire cgroup
146 * set for a task.
147 */
148 struct css_set {
149 /* Reference count */
150 atomic_t refcount;
151
152 /*
153 * List running through all cgroup groups in the same hash
154 * slot. Protected by css_set_lock
155 */
156 struct hlist_node hlist;
157
158 /*
159 * Lists running through all tasks using this cgroup group.
160 * mg_tasks lists tasks which belong to this cset but are in the
161 * process of being migrated out or in. Protected by
162 * css_set_rwsem, but, during migration, once tasks are moved to
163 * mg_tasks, it can be read safely while holding cgroup_mutex.
164 */
165 struct list_head tasks;
166 struct list_head mg_tasks;
167
168 /*
169 * List of cgrp_cset_links pointing at cgroups referenced from this
170 * css_set. Protected by css_set_lock.
171 */
172 struct list_head cgrp_links;
173
174 /* the default cgroup associated with this css_set */
175 struct cgroup *dfl_cgrp;
176
177 /*
178 * Set of subsystem states, one for each subsystem. This array is
179 * immutable after creation apart from the init_css_set during
180 * subsystem registration (at boot time).
181 */
182 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
183
184 /*
185 * List of csets participating in the on-going migration either as
186 * source or destination. Protected by cgroup_mutex.
187 */
188 struct list_head mg_preload_node;
189 struct list_head mg_node;
190
191 /*
192 * If this cset is acting as the source of migration the following
193 * two fields are set. mg_src_cgrp is the source cgroup of the
194 * on-going migration and mg_dst_cset is the destination cset the
195 * target tasks on this cset should be migrated to. Protected by
196 * cgroup_mutex.
197 */
198 struct cgroup *mg_src_cgrp;
199 struct css_set *mg_dst_cset;
200
201 /*
202 * On the default hierarhcy, ->subsys[ssid] may point to a css
203 * attached to an ancestor instead of the cgroup this css_set is
204 * associated with. The following node is anchored at
205 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
206 * iterate through all css's attached to a given cgroup.
207 */
208 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
209
210 /* all css_task_iters currently walking this cset */
211 struct list_head task_iters;
212
213 /* For RCU-protected deletion */
214 struct rcu_head rcu_head;
215 };
216
217 struct cgroup {
218 /* self css with NULL ->ss, points back to this cgroup */
219 struct cgroup_subsys_state self;
220
221 unsigned long flags; /* "unsigned long" so bitops work */
222
223 /*
224 * idr allocated in-hierarchy ID.
225 *
226 * ID 0 is not used, the ID of the root cgroup is always 1, and a
227 * new cgroup will be assigned with a smallest available ID.
228 *
229 * Allocating/Removing ID must be protected by cgroup_mutex.
230 */
231 int id;
232
233 /*
234 * The depth this cgroup is at. The root is at depth zero and each
235 * step down the hierarchy increments the level. This along with
236 * ancestor_ids[] can determine whether a given cgroup is a
237 * descendant of another without traversing the hierarchy.
238 */
239 int level;
240
241 /*
242 * Each non-empty css_set associated with this cgroup contributes
243 * one to populated_cnt. All children with non-zero popuplated_cnt
244 * of their own contribute one. The count is zero iff there's no
245 * task in this cgroup or its subtree.
246 */
247 int populated_cnt;
248
249 struct kernfs_node *kn; /* cgroup kernfs entry */
250 struct cgroup_file procs_file; /* handle for "cgroup.procs" */
251 struct cgroup_file events_file; /* handle for "cgroup.events" */
252
253 /*
254 * The bitmask of subsystems enabled on the child cgroups.
255 * ->subtree_control is the one configured through
256 * "cgroup.subtree_control" while ->child_ss_mask is the effective
257 * one which may have more subsystems enabled. Controller knobs
258 * are made available iff it's enabled in ->subtree_control.
259 */
260 unsigned long subtree_control;
261 unsigned long subtree_ss_mask;
262
263 /* Private pointers for each registered subsystem */
264 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
265
266 struct cgroup_root *root;
267
268 /*
269 * List of cgrp_cset_links pointing at css_sets with tasks in this
270 * cgroup. Protected by css_set_lock.
271 */
272 struct list_head cset_links;
273
274 /*
275 * On the default hierarchy, a css_set for a cgroup with some
276 * susbsys disabled will point to css's which are associated with
277 * the closest ancestor which has the subsys enabled. The
278 * following lists all css_sets which point to this cgroup's css
279 * for the given subsystem.
280 */
281 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
282
283 /*
284 * list of pidlists, up to two for each namespace (one for procs, one
285 * for tasks); created on demand.
286 */
287 struct list_head pidlists;
288 struct mutex pidlist_mutex;
289
290 /* used to wait for offlining of csses */
291 wait_queue_head_t offline_waitq;
292
293 /* used to schedule release agent */
294 struct work_struct release_agent_work;
295
296 /* ids of the ancestors at each level including self */
297 int ancestor_ids[];
298 };
299
300 /*
301 * A cgroup_root represents the root of a cgroup hierarchy, and may be
302 * associated with a kernfs_root to form an active hierarchy. This is
303 * internal to cgroup core. Don't access directly from controllers.
304 */
305 struct cgroup_root {
306 struct kernfs_root *kf_root;
307
308 /* The bitmask of subsystems attached to this hierarchy */
309 unsigned int subsys_mask;
310
311 /* Unique id for this hierarchy. */
312 int hierarchy_id;
313
314 /* The root cgroup. Root is destroyed on its release. */
315 struct cgroup cgrp;
316
317 /* for cgrp->ancestor_ids[0] */
318 int cgrp_ancestor_id_storage;
319
320 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
321 atomic_t nr_cgrps;
322
323 /* A list running through the active hierarchies */
324 struct list_head root_list;
325
326 /* Hierarchy-specific flags */
327 unsigned int flags;
328
329 /* IDs for cgroups in this hierarchy */
330 struct idr cgroup_idr;
331
332 /* The path to use for release notifications. */
333 char release_agent_path[PATH_MAX];
334
335 /* The name for this hierarchy - may be empty */
336 char name[MAX_CGROUP_ROOT_NAMELEN];
337 };
338
339 /*
340 * struct cftype: handler definitions for cgroup control files
341 *
342 * When reading/writing to a file:
343 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
344 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata
345 */
346 struct cftype {
347 /*
348 * By convention, the name should begin with the name of the
349 * subsystem, followed by a period. Zero length string indicates
350 * end of cftype array.
351 */
352 char name[MAX_CFTYPE_NAME];
353 unsigned long private;
354
355 /*
356 * The maximum length of string, excluding trailing nul, that can
357 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
358 */
359 size_t max_write_len;
360
361 /* CFTYPE_* flags */
362 unsigned int flags;
363
364 /*
365 * If non-zero, should contain the offset from the start of css to
366 * a struct cgroup_file field. cgroup will record the handle of
367 * the created file into it. The recorded handle can be used as
368 * long as the containing css remains accessible.
369 */
370 unsigned int file_offset;
371
372 /*
373 * Fields used for internal bookkeeping. Initialized automatically
374 * during registration.
375 */
376 struct cgroup_subsys *ss; /* NULL for cgroup core files */
377 struct list_head node; /* anchored at ss->cfts */
378 struct kernfs_ops *kf_ops;
379
380 /*
381 * read_u64() is a shortcut for the common case of returning a
382 * single integer. Use it in place of read()
383 */
384 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
385 /*
386 * read_s64() is a signed version of read_u64()
387 */
388 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
389
390 /* generic seq_file read interface */
391 int (*seq_show)(struct seq_file *sf, void *v);
392
393 /* optional ops, implement all or none */
394 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
395 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
396 void (*seq_stop)(struct seq_file *sf, void *v);
397
398 /*
399 * write_u64() is a shortcut for the common case of accepting
400 * a single integer (as parsed by simple_strtoull) from
401 * userspace. Use in place of write(); return 0 or error.
402 */
403 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
404 u64 val);
405 /*
406 * write_s64() is a signed version of write_u64()
407 */
408 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
409 s64 val);
410
411 /*
412 * write() is the generic write callback which maps directly to
413 * kernfs write operation and overrides all other operations.
414 * Maximum write size is determined by ->max_write_len. Use
415 * of_css/cft() to access the associated css and cft.
416 */
417 ssize_t (*write)(struct kernfs_open_file *of,
418 char *buf, size_t nbytes, loff_t off);
419
420 #ifdef CONFIG_DEBUG_LOCK_ALLOC
421 struct lock_class_key lockdep_key;
422 #endif
423 };
424
425 /*
426 * Control Group subsystem type.
427 * See Documentation/cgroups/cgroups.txt for details
428 */
429 struct cgroup_subsys {
430 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
431 int (*css_online)(struct cgroup_subsys_state *css);
432 void (*css_offline)(struct cgroup_subsys_state *css);
433 void (*css_released)(struct cgroup_subsys_state *css);
434 void (*css_free)(struct cgroup_subsys_state *css);
435 void (*css_reset)(struct cgroup_subsys_state *css);
436
437 int (*can_attach)(struct cgroup_taskset *tset);
438 void (*cancel_attach)(struct cgroup_taskset *tset);
439 void (*attach)(struct cgroup_taskset *tset);
440 int (*can_fork)(struct task_struct *task);
441 void (*cancel_fork)(struct task_struct *task);
442 void (*fork)(struct task_struct *task);
443 void (*exit)(struct task_struct *task);
444 void (*free)(struct task_struct *task);
445 void (*bind)(struct cgroup_subsys_state *root_css);
446
447 int early_init;
448
449 /*
450 * If %false, this subsystem is properly hierarchical -
451 * configuration, resource accounting and restriction on a parent
452 * cgroup cover those of its children. If %true, hierarchy support
453 * is broken in some ways - some subsystems ignore hierarchy
454 * completely while others are only implemented half-way.
455 *
456 * It's now disallowed to create nested cgroups if the subsystem is
457 * broken and cgroup core will emit a warning message on such
458 * cases. Eventually, all subsystems will be made properly
459 * hierarchical and this will go away.
460 */
461 bool broken_hierarchy;
462 bool warned_broken_hierarchy;
463
464 /* the following two fields are initialized automtically during boot */
465 int id;
466 const char *name;
467
468 /* optional, initialized automatically during boot if not set */
469 const char *legacy_name;
470
471 /* link to parent, protected by cgroup_lock() */
472 struct cgroup_root *root;
473
474 /* idr for css->id */
475 struct idr css_idr;
476
477 /*
478 * List of cftypes. Each entry is the first entry of an array
479 * terminated by zero length name.
480 */
481 struct list_head cfts;
482
483 /*
484 * Base cftypes which are automatically registered. The two can
485 * point to the same array.
486 */
487 struct cftype *dfl_cftypes; /* for the default hierarchy */
488 struct cftype *legacy_cftypes; /* for the legacy hierarchies */
489
490 /*
491 * A subsystem may depend on other subsystems. When such subsystem
492 * is enabled on a cgroup, the depended-upon subsystems are enabled
493 * together if available. Subsystems enabled due to dependency are
494 * not visible to userland until explicitly enabled. The following
495 * specifies the mask of subsystems that this one depends on.
496 */
497 unsigned int depends_on;
498 };
499
500 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
501
502 /**
503 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
504 * @tsk: target task
505 *
506 * Called from threadgroup_change_begin() and allows cgroup operations to
507 * synchronize against threadgroup changes using a percpu_rw_semaphore.
508 */
509 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
510 {
511 percpu_down_read(&cgroup_threadgroup_rwsem);
512 }
513
514 /**
515 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
516 * @tsk: target task
517 *
518 * Called from threadgroup_change_end(). Counterpart of
519 * cgroup_threadcgroup_change_begin().
520 */
521 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
522 {
523 percpu_up_read(&cgroup_threadgroup_rwsem);
524 }
525
526 #else /* CONFIG_CGROUPS */
527
528 #define CGROUP_SUBSYS_COUNT 0
529
530 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) {}
531 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
532
533 #endif /* CONFIG_CGROUPS */
534
535 #ifdef CONFIG_SOCK_CGROUP_DATA
536
537 /*
538 * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains
539 * per-socket cgroup information except for memcg association.
540 *
541 * On legacy hierarchies, net_prio and net_cls controllers directly set
542 * attributes on each sock which can then be tested by the network layer.
543 * On the default hierarchy, each sock is associated with the cgroup it was
544 * created in and the networking layer can match the cgroup directly.
545 *
546 * To avoid carrying all three cgroup related fields separately in sock,
547 * sock_cgroup_data overloads (prioidx, classid) and the cgroup pointer.
548 * On boot, sock_cgroup_data records the cgroup that the sock was created
549 * in so that cgroup2 matches can be made; however, once either net_prio or
550 * net_cls starts being used, the area is overriden to carry prioidx and/or
551 * classid. The two modes are distinguished by whether the lowest bit is
552 * set. Clear bit indicates cgroup pointer while set bit prioidx and
553 * classid.
554 *
555 * While userland may start using net_prio or net_cls at any time, once
556 * either is used, cgroup2 matching no longer works. There is no reason to
557 * mix the two and this is in line with how legacy and v2 compatibility is
558 * handled. On mode switch, cgroup references which are already being
559 * pointed to by socks may be leaked. While this can be remedied by adding
560 * synchronization around sock_cgroup_data, given that the number of leaked
561 * cgroups is bound and highly unlikely to be high, this seems to be the
562 * better trade-off.
563 */
564 struct sock_cgroup_data {
565 union {
566 #ifdef __LITTLE_ENDIAN
567 struct {
568 u8 is_data;
569 u8 padding;
570 u16 prioidx;
571 u32 classid;
572 } __packed;
573 #else
574 struct {
575 u32 classid;
576 u16 prioidx;
577 u8 padding;
578 u8 is_data;
579 } __packed;
580 #endif
581 u64 val;
582 };
583 };
584
585 /*
586 * There's a theoretical window where the following accessors race with
587 * updaters and return part of the previous pointer as the prioidx or
588 * classid. Such races are short-lived and the result isn't critical.
589 */
590 static inline u16 sock_cgroup_prioidx(struct sock_cgroup_data *skcd)
591 {
592 /* fallback to 1 which is always the ID of the root cgroup */
593 return (skcd->is_data & 1) ? skcd->prioidx : 1;
594 }
595
596 static inline u32 sock_cgroup_classid(struct sock_cgroup_data *skcd)
597 {
598 /* fallback to 0 which is the unconfigured default classid */
599 return (skcd->is_data & 1) ? skcd->classid : 0;
600 }
601
602 /*
603 * If invoked concurrently, the updaters may clobber each other. The
604 * caller is responsible for synchronization.
605 */
606 static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd,
607 u16 prioidx)
608 {
609 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
610
611 if (sock_cgroup_prioidx(&skcd_buf) == prioidx)
612 return;
613
614 if (!(skcd_buf.is_data & 1)) {
615 skcd_buf.val = 0;
616 skcd_buf.is_data = 1;
617 }
618
619 skcd_buf.prioidx = prioidx;
620 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
621 }
622
623 static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd,
624 u32 classid)
625 {
626 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
627
628 if (sock_cgroup_classid(&skcd_buf) == classid)
629 return;
630
631 if (!(skcd_buf.is_data & 1)) {
632 skcd_buf.val = 0;
633 skcd_buf.is_data = 1;
634 }
635
636 skcd_buf.classid = classid;
637 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
638 }
639
640 #else /* CONFIG_SOCK_CGROUP_DATA */
641
642 struct sock_cgroup_data {
643 };
644
645 #endif /* CONFIG_SOCK_CGROUP_DATA */
646
647 #endif /* _LINUX_CGROUP_DEFS_H */