1 #ifndef _LINUX_CGROUP_H
2 #define _LINUX_CGROUP_H
3 /*
4 * cgroup interface
5 *
6 * Copyright (C) 2003 BULL SA
7 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
8 *
9 */
10
11 #include <linux/sched.h>
12 #include <linux/cpumask.h>
13 #include <linux/nodemask.h>
14 #include <linux/rcupdate.h>
15 #include <linux/rculist.h>
16 #include <linux/cgroupstats.h>
17 #include <linux/rwsem.h>
18 #include <linux/idr.h>
19 #include <linux/workqueue.h>
20 #include <linux/fs.h>
21 #include <linux/percpu-refcount.h>
22 #include <linux/seq_file.h>
23 #include <linux/kernfs.h>
24 #include <linux/wait.h>
25
26 #ifdef CONFIG_CGROUPS
27
28 struct cgroup_root;
29 struct cgroup_subsys;
30 struct cgroup;
31
32 extern int cgroup_init_early(void);
33 extern int cgroup_init(void);
34 extern void cgroup_fork(struct task_struct *p);
35 extern void cgroup_post_fork(struct task_struct *p);
36 extern void cgroup_exit(struct task_struct *p);
37 extern int cgroupstats_build(struct cgroupstats *stats,
38 struct dentry *dentry);
39
40 extern int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
41 struct pid *pid, struct task_struct *tsk);
42
43 /* define the enumeration of all cgroup subsystems */
44 #define SUBSYS(_x) _x ## _cgrp_id,
45 enum cgroup_subsys_id {
46 #include <linux/cgroup_subsys.h>
47 CGROUP_SUBSYS_COUNT,
48 };
49 #undef SUBSYS
50
51 /*
52 * Per-subsystem/per-cgroup state maintained by the system. This is the
53 * fundamental structural building block that controllers deal with.
54 *
55 * Fields marked with "PI:" are public and immutable and may be accessed
56 * directly without synchronization.
57 */
58 struct cgroup_subsys_state {
59 /* PI: the cgroup that this css is attached to */
60 struct cgroup *cgroup;
61
62 /* PI: the cgroup subsystem that this css is attached to */
63 struct cgroup_subsys *ss;
64
65 /* reference count - access via css_[try]get() and css_put() */
66 struct percpu_ref refcnt;
67
68 /* PI: the parent css */
69 struct cgroup_subsys_state *parent;
70
71 /* siblings list anchored at the parent's ->children */
72 struct list_head sibling;
73 struct list_head children;
74
75 /*
76 * PI: Subsys-unique ID. 0 is unused and root is always 1. The
77 * matching css can be looked up using css_from_id().
78 */
79 int id;
80
81 unsigned int flags;
82
83 /*
84 * Monotonically increasing unique serial number which defines a
85 * uniform order among all csses. It's guaranteed that all
86 * ->children lists are in the ascending order of ->serial_nr and
87 * used to allow interrupting and resuming iterations.
88 */
89 u64 serial_nr;
90
91 /* percpu_ref killing and RCU release */
92 struct rcu_head rcu_head;
93 struct work_struct destroy_work;
94 };
95
96 /* bits in struct cgroup_subsys_state flags field */
97 enum {
98 CSS_NO_REF = (1 << 0), /* no reference counting for this css */
99 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
100 CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */
101 };
102
103 /**
104 * css_get - obtain a reference on the specified css
105 * @css: target css
106 *
107 * The caller must already have a reference.
108 */
css_get(struct cgroup_subsys_state * css)109 static inline void css_get(struct cgroup_subsys_state *css)
110 {
111 if (!(css->flags & CSS_NO_REF))
112 percpu_ref_get(&css->refcnt);
113 }
114
115 /**
116 * css_tryget - try to obtain a reference on the specified css
117 * @css: target css
118 *
119 * Obtain a reference on @css unless it already has reached zero and is
120 * being released. This function doesn't care whether @css is on or
121 * offline. The caller naturally needs to ensure that @css is accessible
122 * but doesn't have to be holding a reference on it - IOW, RCU protected
123 * access is good enough for this function. Returns %true if a reference
124 * count was successfully obtained; %false otherwise.
125 */
css_tryget(struct cgroup_subsys_state * css)126 static inline bool css_tryget(struct cgroup_subsys_state *css)
127 {
128 if (!(css->flags & CSS_NO_REF))
129 return percpu_ref_tryget(&css->refcnt);
130 return true;
131 }
132
133 /**
134 * css_tryget_online - try to obtain a reference on the specified css if online
135 * @css: target css
136 *
137 * Obtain a reference on @css if it's online. The caller naturally needs
138 * to ensure that @css is accessible but doesn't have to be holding a
139 * reference on it - IOW, RCU protected access is good enough for this
140 * function. Returns %true if a reference count was successfully obtained;
141 * %false otherwise.
142 */
css_tryget_online(struct cgroup_subsys_state * css)143 static inline bool css_tryget_online(struct cgroup_subsys_state *css)
144 {
145 if (!(css->flags & CSS_NO_REF))
146 return percpu_ref_tryget_live(&css->refcnt);
147 return true;
148 }
149
150 /**
151 * css_put - put a css reference
152 * @css: target css
153 *
154 * Put a reference obtained via css_get() and css_tryget_online().
155 */
css_put(struct cgroup_subsys_state * css)156 static inline void css_put(struct cgroup_subsys_state *css)
157 {
158 if (!(css->flags & CSS_NO_REF))
159 percpu_ref_put(&css->refcnt);
160 }
161
162 /* bits in struct cgroup flags field */
163 enum {
164 /* Control Group requires release notifications to userspace */
165 CGRP_NOTIFY_ON_RELEASE,
166 /*
167 * Clone the parent's configuration when creating a new child
168 * cpuset cgroup. For historical reasons, this option can be
169 * specified at mount time and thus is implemented here.
170 */
171 CGRP_CPUSET_CLONE_CHILDREN,
172 };
173
174 struct cgroup {
175 /* self css with NULL ->ss, points back to this cgroup */
176 struct cgroup_subsys_state self;
177
178 unsigned long flags; /* "unsigned long" so bitops work */
179
180 /*
181 * idr allocated in-hierarchy ID.
182 *
183 * ID 0 is not used, the ID of the root cgroup is always 1, and a
184 * new cgroup will be assigned with a smallest available ID.
185 *
186 * Allocating/Removing ID must be protected by cgroup_mutex.
187 */
188 int id;
189
190 /*
191 * If this cgroup contains any tasks, it contributes one to
192 * populated_cnt. All children with non-zero popuplated_cnt of
193 * their own contribute one. The count is zero iff there's no task
194 * in this cgroup or its subtree.
195 */
196 int populated_cnt;
197
198 struct kernfs_node *kn; /* cgroup kernfs entry */
199 struct kernfs_node *populated_kn; /* kn for "cgroup.subtree_populated" */
200
201 /*
202 * The bitmask of subsystems enabled on the child cgroups.
203 * ->subtree_control is the one configured through
204 * "cgroup.subtree_control" while ->child_subsys_mask is the
205 * effective one which may have more subsystems enabled.
206 * Controller knobs are made available iff it's enabled in
207 * ->subtree_control.
208 */
209 unsigned int subtree_control;
210 unsigned int child_subsys_mask;
211
212 /* Private pointers for each registered subsystem */
213 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
214
215 struct cgroup_root *root;
216
217 /*
218 * List of cgrp_cset_links pointing at css_sets with tasks in this
219 * cgroup. Protected by css_set_lock.
220 */
221 struct list_head cset_links;
222
223 /*
224 * On the default hierarchy, a css_set for a cgroup with some
225 * susbsys disabled will point to css's which are associated with
226 * the closest ancestor which has the subsys enabled. The
227 * following lists all css_sets which point to this cgroup's css
228 * for the given subsystem.
229 */
230 struct list_head e_csets[CGROUP_SUBSYS_COUNT];
231
232 /*
233 * list of pidlists, up to two for each namespace (one for procs, one
234 * for tasks); created on demand.
235 */
236 struct list_head pidlists;
237 struct mutex pidlist_mutex;
238
239 /* used to wait for offlining of csses */
240 wait_queue_head_t offline_waitq;
241
242 /* used to schedule release agent */
243 struct work_struct release_agent_work;
244 };
245
246 #define MAX_CGROUP_ROOT_NAMELEN 64
247
248 /* cgroup_root->flags */
249 enum {
250 CGRP_ROOT_SANE_BEHAVIOR = (1 << 0), /* __DEVEL__sane_behavior specified */
251 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
252 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
253 };
254
255 /*
256 * A cgroup_root represents the root of a cgroup hierarchy, and may be
257 * associated with a kernfs_root to form an active hierarchy. This is
258 * internal to cgroup core. Don't access directly from controllers.
259 */
260 struct cgroup_root {
261 struct kernfs_root *kf_root;
262
263 /* The bitmask of subsystems attached to this hierarchy */
264 unsigned int subsys_mask;
265
266 /* Unique id for this hierarchy. */
267 int hierarchy_id;
268
269 /* The root cgroup. Root is destroyed on its release. */
270 struct cgroup cgrp;
271
272 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
273 atomic_t nr_cgrps;
274
275 /* A list running through the active hierarchies */
276 struct list_head root_list;
277
278 /* Hierarchy-specific flags */
279 unsigned int flags;
280
281 /* IDs for cgroups in this hierarchy */
282 struct idr cgroup_idr;
283
284 /* The path to use for release notifications. */
285 char release_agent_path[PATH_MAX];
286
287 /* The name for this hierarchy - may be empty */
288 char name[MAX_CGROUP_ROOT_NAMELEN];
289 };
290
291 /*
292 * A css_set is a structure holding pointers to a set of
293 * cgroup_subsys_state objects. This saves space in the task struct
294 * object and speeds up fork()/exit(), since a single inc/dec and a
295 * list_add()/del() can bump the reference count on the entire cgroup
296 * set for a task.
297 */
298
299 struct css_set {
300
301 /* Reference count */
302 atomic_t refcount;
303
304 /*
305 * List running through all cgroup groups in the same hash
306 * slot. Protected by css_set_lock
307 */
308 struct hlist_node hlist;
309
310 /*
311 * Lists running through all tasks using this cgroup group.
312 * mg_tasks lists tasks which belong to this cset but are in the
313 * process of being migrated out or in. Protected by
314 * css_set_rwsem, but, during migration, once tasks are moved to
315 * mg_tasks, it can be read safely while holding cgroup_mutex.
316 */
317 struct list_head tasks;
318 struct list_head mg_tasks;
319
320 /*
321 * List of cgrp_cset_links pointing at cgroups referenced from this
322 * css_set. Protected by css_set_lock.
323 */
324 struct list_head cgrp_links;
325
326 /* the default cgroup associated with this css_set */
327 struct cgroup *dfl_cgrp;
328
329 /*
330 * Set of subsystem states, one for each subsystem. This array is
331 * immutable after creation apart from the init_css_set during
332 * subsystem registration (at boot time).
333 */
334 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
335
336 /*
337 * List of csets participating in the on-going migration either as
338 * source or destination. Protected by cgroup_mutex.
339 */
340 struct list_head mg_preload_node;
341 struct list_head mg_node;
342
343 /*
344 * If this cset is acting as the source of migration the following
345 * two fields are set. mg_src_cgrp is the source cgroup of the
346 * on-going migration and mg_dst_cset is the destination cset the
347 * target tasks on this cset should be migrated to. Protected by
348 * cgroup_mutex.
349 */
350 struct cgroup *mg_src_cgrp;
351 struct css_set *mg_dst_cset;
352
353 /*
354 * On the default hierarhcy, ->subsys[ssid] may point to a css
355 * attached to an ancestor instead of the cgroup this css_set is
356 * associated with. The following node is anchored at
357 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
358 * iterate through all css's attached to a given cgroup.
359 */
360 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
361
362 /* For RCU-protected deletion */
363 struct rcu_head rcu_head;
364 };
365
366 /*
367 * struct cftype: handler definitions for cgroup control files
368 *
369 * When reading/writing to a file:
370 * - the cgroup to use is file->f_dentry->d_parent->d_fsdata
371 * - the 'cftype' of the file is file->f_dentry->d_fsdata
372 */
373
374 /* cftype->flags */
375 enum {
376 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
377 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
378 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
379
380 /* internal flags, do not use outside cgroup core proper */
381 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */
382 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */
383 };
384
385 #define MAX_CFTYPE_NAME 64
386
387 struct cftype {
388 /*
389 * By convention, the name should begin with the name of the
390 * subsystem, followed by a period. Zero length string indicates
391 * end of cftype array.
392 */
393 char name[MAX_CFTYPE_NAME];
394 int private;
395 /*
396 * If not 0, file mode is set to this value, otherwise it will
397 * be figured out automatically
398 */
399 umode_t mode;
400
401 /*
402 * The maximum length of string, excluding trailing nul, that can
403 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
404 */
405 size_t max_write_len;
406
407 /* CFTYPE_* flags */
408 unsigned int flags;
409
410 /*
411 * Fields used for internal bookkeeping. Initialized automatically
412 * during registration.
413 */
414 struct cgroup_subsys *ss; /* NULL for cgroup core files */
415 struct list_head node; /* anchored at ss->cfts */
416 struct kernfs_ops *kf_ops;
417
418 /*
419 * read_u64() is a shortcut for the common case of returning a
420 * single integer. Use it in place of read()
421 */
422 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
423 /*
424 * read_s64() is a signed version of read_u64()
425 */
426 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
427
428 /* generic seq_file read interface */
429 int (*seq_show)(struct seq_file *sf, void *v);
430
431 /* optional ops, implement all or none */
432 void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
433 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
434 void (*seq_stop)(struct seq_file *sf, void *v);
435
436 /*
437 * write_u64() is a shortcut for the common case of accepting
438 * a single integer (as parsed by simple_strtoull) from
439 * userspace. Use in place of write(); return 0 or error.
440 */
441 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
442 u64 val);
443 /*
444 * write_s64() is a signed version of write_u64()
445 */
446 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
447 s64 val);
448
449 /*
450 * write() is the generic write callback which maps directly to
451 * kernfs write operation and overrides all other operations.
452 * Maximum write size is determined by ->max_write_len. Use
453 * of_css/cft() to access the associated css and cft.
454 */
455 ssize_t (*write)(struct kernfs_open_file *of,
456 char *buf, size_t nbytes, loff_t off);
457
458 #ifdef CONFIG_DEBUG_LOCK_ALLOC
459 struct lock_class_key lockdep_key;
460 #endif
461 };
462
463 extern struct cgroup_root cgrp_dfl_root;
464 extern struct css_set init_css_set;
465
466 /**
467 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
468 * @cgrp: the cgroup of interest
469 *
470 * The default hierarchy is the v2 interface of cgroup and this function
471 * can be used to test whether a cgroup is on the default hierarchy for
472 * cases where a subsystem should behave differnetly depending on the
473 * interface version.
474 *
475 * The set of behaviors which change on the default hierarchy are still
476 * being determined and the mount option is prefixed with __DEVEL__.
477 *
478 * List of changed behaviors:
479 *
480 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
481 * and "name" are disallowed.
482 *
483 * - When mounting an existing superblock, mount options should match.
484 *
485 * - Remount is disallowed.
486 *
487 * - rename(2) is disallowed.
488 *
489 * - "tasks" is removed. Everything should be at process granularity. Use
490 * "cgroup.procs" instead.
491 *
492 * - "cgroup.procs" is not sorted. pids will be unique unless they got
493 * recycled inbetween reads.
494 *
495 * - "release_agent" and "notify_on_release" are removed. Replacement
496 * notification mechanism will be implemented.
497 *
498 * - "cgroup.clone_children" is removed.
499 *
500 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
501 * and its descendants contain no task; otherwise, 1. The file also
502 * generates kernfs notification which can be monitored through poll and
503 * [di]notify when the value of the file changes.
504 *
505 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
506 * take masks of ancestors with non-empty cpus/mems, instead of being
507 * moved to an ancestor.
508 *
509 * - cpuset: a task can be moved into an empty cpuset, and again it takes
510 * masks of ancestors.
511 *
512 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
513 * is not created.
514 *
515 * - blkcg: blk-throttle becomes properly hierarchical.
516 *
517 * - debug: disallowed on the default hierarchy.
518 */
cgroup_on_dfl(const struct cgroup * cgrp)519 static inline bool cgroup_on_dfl(const struct cgroup *cgrp)
520 {
521 return cgrp->root == &cgrp_dfl_root;
522 }
523
524 /* no synchronization, the result can only be used as a hint */
cgroup_has_tasks(struct cgroup * cgrp)525 static inline bool cgroup_has_tasks(struct cgroup *cgrp)
526 {
527 return !list_empty(&cgrp->cset_links);
528 }
529
530 /* returns ino associated with a cgroup */
cgroup_ino(struct cgroup * cgrp)531 static inline ino_t cgroup_ino(struct cgroup *cgrp)
532 {
533 return cgrp->kn->ino;
534 }
535
536 /* cft/css accessors for cftype->write() operation */
of_cft(struct kernfs_open_file * of)537 static inline struct cftype *of_cft(struct kernfs_open_file *of)
538 {
539 return of->kn->priv;
540 }
541
542 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of);
543
544 /* cft/css accessors for cftype->seq_*() operations */
seq_cft(struct seq_file * seq)545 static inline struct cftype *seq_cft(struct seq_file *seq)
546 {
547 return of_cft(seq->private);
548 }
549
seq_css(struct seq_file * seq)550 static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq)
551 {
552 return of_css(seq->private);
553 }
554
555 /*
556 * Name / path handling functions. All are thin wrappers around the kernfs
557 * counterparts and can be called under any context.
558 */
559
cgroup_name(struct cgroup * cgrp,char * buf,size_t buflen)560 static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen)
561 {
562 return kernfs_name(cgrp->kn, buf, buflen);
563 }
564
cgroup_path(struct cgroup * cgrp,char * buf,size_t buflen)565 static inline char * __must_check cgroup_path(struct cgroup *cgrp, char *buf,
566 size_t buflen)
567 {
568 return kernfs_path(cgrp->kn, buf, buflen);
569 }
570
pr_cont_cgroup_name(struct cgroup * cgrp)571 static inline void pr_cont_cgroup_name(struct cgroup *cgrp)
572 {
573 pr_cont_kernfs_name(cgrp->kn);
574 }
575
pr_cont_cgroup_path(struct cgroup * cgrp)576 static inline void pr_cont_cgroup_path(struct cgroup *cgrp)
577 {
578 pr_cont_kernfs_path(cgrp->kn);
579 }
580
581 char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen);
582
583 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
584 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts);
585 int cgroup_rm_cftypes(struct cftype *cfts);
586
587 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor);
588
589 /*
590 * Control Group taskset, used to pass around set of tasks to cgroup_subsys
591 * methods.
592 */
593 struct cgroup_taskset;
594 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset);
595 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset);
596
597 /**
598 * cgroup_taskset_for_each - iterate cgroup_taskset
599 * @task: the loop cursor
600 * @tset: taskset to iterate
601 */
602 #define cgroup_taskset_for_each(task, tset) \
603 for ((task) = cgroup_taskset_first((tset)); (task); \
604 (task) = cgroup_taskset_next((tset)))
605
606 /*
607 * Control Group subsystem type.
608 * See Documentation/cgroups/cgroups.txt for details
609 */
610
611 struct cgroup_subsys {
612 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
613 int (*css_online)(struct cgroup_subsys_state *css);
614 void (*css_offline)(struct cgroup_subsys_state *css);
615 void (*css_free)(struct cgroup_subsys_state *css);
616 void (*css_reset)(struct cgroup_subsys_state *css);
617
618 int (*can_attach)(struct cgroup_subsys_state *css,
619 struct cgroup_taskset *tset);
620 void (*cancel_attach)(struct cgroup_subsys_state *css,
621 struct cgroup_taskset *tset);
622 void (*attach)(struct cgroup_subsys_state *css,
623 struct cgroup_taskset *tset);
624 void (*fork)(struct task_struct *task);
625 void (*exit)(struct cgroup_subsys_state *css,
626 struct cgroup_subsys_state *old_css,
627 struct task_struct *task);
628 void (*bind)(struct cgroup_subsys_state *root_css);
629
630 int disabled;
631 int early_init;
632
633 /*
634 * If %false, this subsystem is properly hierarchical -
635 * configuration, resource accounting and restriction on a parent
636 * cgroup cover those of its children. If %true, hierarchy support
637 * is broken in some ways - some subsystems ignore hierarchy
638 * completely while others are only implemented half-way.
639 *
640 * It's now disallowed to create nested cgroups if the subsystem is
641 * broken and cgroup core will emit a warning message on such
642 * cases. Eventually, all subsystems will be made properly
643 * hierarchical and this will go away.
644 */
645 bool broken_hierarchy;
646 bool warned_broken_hierarchy;
647
648 /* the following two fields are initialized automtically during boot */
649 int id;
650 #define MAX_CGROUP_TYPE_NAMELEN 32
651 const char *name;
652
653 /* link to parent, protected by cgroup_lock() */
654 struct cgroup_root *root;
655
656 /* idr for css->id */
657 struct idr css_idr;
658
659 /*
660 * List of cftypes. Each entry is the first entry of an array
661 * terminated by zero length name.
662 */
663 struct list_head cfts;
664
665 /*
666 * Base cftypes which are automatically registered. The two can
667 * point to the same array.
668 */
669 struct cftype *dfl_cftypes; /* for the default hierarchy */
670 struct cftype *legacy_cftypes; /* for the legacy hierarchies */
671
672 /*
673 * A subsystem may depend on other subsystems. When such subsystem
674 * is enabled on a cgroup, the depended-upon subsystems are enabled
675 * together if available. Subsystems enabled due to dependency are
676 * not visible to userland until explicitly enabled. The following
677 * specifies the mask of subsystems that this one depends on.
678 */
679 unsigned int depends_on;
680 };
681
682 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys;
683 #include <linux/cgroup_subsys.h>
684 #undef SUBSYS
685
686 /**
687 * task_css_set_check - obtain a task's css_set with extra access conditions
688 * @task: the task to obtain css_set for
689 * @__c: extra condition expression to be passed to rcu_dereference_check()
690 *
691 * A task's css_set is RCU protected, initialized and exited while holding
692 * task_lock(), and can only be modified while holding both cgroup_mutex
693 * and task_lock() while the task is alive. This macro verifies that the
694 * caller is inside proper critical section and returns @task's css_set.
695 *
696 * The caller can also specify additional allowed conditions via @__c, such
697 * as locks used during the cgroup_subsys::attach() methods.
698 */
699 #ifdef CONFIG_PROVE_RCU
700 extern struct mutex cgroup_mutex;
701 extern struct rw_semaphore css_set_rwsem;
702 #define task_css_set_check(task, __c) \
703 rcu_dereference_check((task)->cgroups, \
704 lockdep_is_held(&cgroup_mutex) || \
705 lockdep_is_held(&css_set_rwsem) || \
706 ((task)->flags & PF_EXITING) || (__c))
707 #else
708 #define task_css_set_check(task, __c) \
709 rcu_dereference((task)->cgroups)
710 #endif
711
712 /**
713 * task_css_check - obtain css for (task, subsys) w/ extra access conds
714 * @task: the target task
715 * @subsys_id: the target subsystem ID
716 * @__c: extra condition expression to be passed to rcu_dereference_check()
717 *
718 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The
719 * synchronization rules are the same as task_css_set_check().
720 */
721 #define task_css_check(task, subsys_id, __c) \
722 task_css_set_check((task), (__c))->subsys[(subsys_id)]
723
724 /**
725 * task_css_set - obtain a task's css_set
726 * @task: the task to obtain css_set for
727 *
728 * See task_css_set_check().
729 */
task_css_set(struct task_struct * task)730 static inline struct css_set *task_css_set(struct task_struct *task)
731 {
732 return task_css_set_check(task, false);
733 }
734
735 /**
736 * task_css - obtain css for (task, subsys)
737 * @task: the target task
738 * @subsys_id: the target subsystem ID
739 *
740 * See task_css_check().
741 */
task_css(struct task_struct * task,int subsys_id)742 static inline struct cgroup_subsys_state *task_css(struct task_struct *task,
743 int subsys_id)
744 {
745 return task_css_check(task, subsys_id, false);
746 }
747
748 /**
749 * task_css_is_root - test whether a task belongs to the root css
750 * @task: the target task
751 * @subsys_id: the target subsystem ID
752 *
753 * Test whether @task belongs to the root css on the specified subsystem.
754 * May be invoked in any context.
755 */
task_css_is_root(struct task_struct * task,int subsys_id)756 static inline bool task_css_is_root(struct task_struct *task, int subsys_id)
757 {
758 return task_css_check(task, subsys_id, true) ==
759 init_css_set.subsys[subsys_id];
760 }
761
task_cgroup(struct task_struct * task,int subsys_id)762 static inline struct cgroup *task_cgroup(struct task_struct *task,
763 int subsys_id)
764 {
765 return task_css(task, subsys_id)->cgroup;
766 }
767
768 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
769 struct cgroup_subsys_state *parent);
770
771 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss);
772
773 /**
774 * css_for_each_child - iterate through children of a css
775 * @pos: the css * to use as the loop cursor
776 * @parent: css whose children to walk
777 *
778 * Walk @parent's children. Must be called under rcu_read_lock().
779 *
780 * If a subsystem synchronizes ->css_online() and the start of iteration, a
781 * css which finished ->css_online() is guaranteed to be visible in the
782 * future iterations and will stay visible until the last reference is put.
783 * A css which hasn't finished ->css_online() or already finished
784 * ->css_offline() may show up during traversal. It's each subsystem's
785 * responsibility to synchronize against on/offlining.
786 *
787 * It is allowed to temporarily drop RCU read lock during iteration. The
788 * caller is responsible for ensuring that @pos remains accessible until
789 * the start of the next iteration by, for example, bumping the css refcnt.
790 */
791 #define css_for_each_child(pos, parent) \
792 for ((pos) = css_next_child(NULL, (parent)); (pos); \
793 (pos) = css_next_child((pos), (parent)))
794
795 struct cgroup_subsys_state *
796 css_next_descendant_pre(struct cgroup_subsys_state *pos,
797 struct cgroup_subsys_state *css);
798
799 struct cgroup_subsys_state *
800 css_rightmost_descendant(struct cgroup_subsys_state *pos);
801
802 /**
803 * css_for_each_descendant_pre - pre-order walk of a css's descendants
804 * @pos: the css * to use as the loop cursor
805 * @root: css whose descendants to walk
806 *
807 * Walk @root's descendants. @root is included in the iteration and the
808 * first node to be visited. Must be called under rcu_read_lock().
809 *
810 * If a subsystem synchronizes ->css_online() and the start of iteration, a
811 * css which finished ->css_online() is guaranteed to be visible in the
812 * future iterations and will stay visible until the last reference is put.
813 * A css which hasn't finished ->css_online() or already finished
814 * ->css_offline() may show up during traversal. It's each subsystem's
815 * responsibility to synchronize against on/offlining.
816 *
817 * For example, the following guarantees that a descendant can't escape
818 * state updates of its ancestors.
819 *
820 * my_online(@css)
821 * {
822 * Lock @css's parent and @css;
823 * Inherit state from the parent;
824 * Unlock both.
825 * }
826 *
827 * my_update_state(@css)
828 * {
829 * css_for_each_descendant_pre(@pos, @css) {
830 * Lock @pos;
831 * if (@pos == @css)
832 * Update @css's state;
833 * else
834 * Verify @pos is alive and inherit state from its parent;
835 * Unlock @pos;
836 * }
837 * }
838 *
839 * As long as the inheriting step, including checking the parent state, is
840 * enclosed inside @pos locking, double-locking the parent isn't necessary
841 * while inheriting. The state update to the parent is guaranteed to be
842 * visible by walking order and, as long as inheriting operations to the
843 * same @pos are atomic to each other, multiple updates racing each other
844 * still result in the correct state. It's guaranateed that at least one
845 * inheritance happens for any css after the latest update to its parent.
846 *
847 * If checking parent's state requires locking the parent, each inheriting
848 * iteration should lock and unlock both @pos->parent and @pos.
849 *
850 * Alternatively, a subsystem may choose to use a single global lock to
851 * synchronize ->css_online() and ->css_offline() against tree-walking
852 * operations.
853 *
854 * It is allowed to temporarily drop RCU read lock during iteration. The
855 * caller is responsible for ensuring that @pos remains accessible until
856 * the start of the next iteration by, for example, bumping the css refcnt.
857 */
858 #define css_for_each_descendant_pre(pos, css) \
859 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \
860 (pos) = css_next_descendant_pre((pos), (css)))
861
862 struct cgroup_subsys_state *
863 css_next_descendant_post(struct cgroup_subsys_state *pos,
864 struct cgroup_subsys_state *css);
865
866 /**
867 * css_for_each_descendant_post - post-order walk of a css's descendants
868 * @pos: the css * to use as the loop cursor
869 * @css: css whose descendants to walk
870 *
871 * Similar to css_for_each_descendant_pre() but performs post-order
872 * traversal instead. @root is included in the iteration and the last
873 * node to be visited.
874 *
875 * If a subsystem synchronizes ->css_online() and the start of iteration, a
876 * css which finished ->css_online() is guaranteed to be visible in the
877 * future iterations and will stay visible until the last reference is put.
878 * A css which hasn't finished ->css_online() or already finished
879 * ->css_offline() may show up during traversal. It's each subsystem's
880 * responsibility to synchronize against on/offlining.
881 *
882 * Note that the walk visibility guarantee example described in pre-order
883 * walk doesn't apply the same to post-order walks.
884 */
885 #define css_for_each_descendant_post(pos, css) \
886 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \
887 (pos) = css_next_descendant_post((pos), (css)))
888
889 bool css_has_online_children(struct cgroup_subsys_state *css);
890
891 /* A css_task_iter should be treated as an opaque object */
892 struct css_task_iter {
893 struct cgroup_subsys *ss;
894
895 struct list_head *cset_pos;
896 struct list_head *cset_head;
897
898 struct list_head *task_pos;
899 struct list_head *tasks_head;
900 struct list_head *mg_tasks_head;
901 };
902
903 void css_task_iter_start(struct cgroup_subsys_state *css,
904 struct css_task_iter *it);
905 struct task_struct *css_task_iter_next(struct css_task_iter *it);
906 void css_task_iter_end(struct css_task_iter *it);
907
908 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *);
909 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from);
910
911 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
912 struct cgroup_subsys *ss);
913
914 #else /* !CONFIG_CGROUPS */
915
cgroup_init_early(void)916 static inline int cgroup_init_early(void) { return 0; }
cgroup_init(void)917 static inline int cgroup_init(void) { return 0; }
cgroup_fork(struct task_struct * p)918 static inline void cgroup_fork(struct task_struct *p) {}
cgroup_post_fork(struct task_struct * p)919 static inline void cgroup_post_fork(struct task_struct *p) {}
cgroup_exit(struct task_struct * p)920 static inline void cgroup_exit(struct task_struct *p) {}
921
cgroupstats_build(struct cgroupstats * stats,struct dentry * dentry)922 static inline int cgroupstats_build(struct cgroupstats *stats,
923 struct dentry *dentry)
924 {
925 return -EINVAL;
926 }
927
928 /* No cgroups - nothing to do */
cgroup_attach_task_all(struct task_struct * from,struct task_struct * t)929 static inline int cgroup_attach_task_all(struct task_struct *from,
930 struct task_struct *t)
931 {
932 return 0;
933 }
934
935 #endif /* !CONFIG_CGROUPS */
936
937 #endif /* _LINUX_CGROUP_H */
938