1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef __CGROUP_INTERNAL_H
3#define __CGROUP_INTERNAL_H
4
5#include <linux/cgroup.h>
6#include <linux/kernfs.h>
7#include <linux/workqueue.h>
8#include <linux/list.h>
9#include <linux/refcount.h>
10#include <linux/fs_parser.h>
11
12#define TRACE_CGROUP_PATH_LEN 1024
13extern spinlock_t trace_cgroup_path_lock;
14extern char trace_cgroup_path[TRACE_CGROUP_PATH_LEN];
15extern void __init enable_debug_cgroup(void);
16
17/*
18 * cgroup_path() takes a spin lock. It is good practice not to take
19 * spin locks within trace point handlers, as they are mostly hidden
20 * from normal view. As cgroup_path() can take the kernfs_rename_lock
21 * spin lock, it is best to not call that function from the trace event
22 * handler.
23 *
24 * Note: trace_cgroup_##type##_enabled() is a static branch that will only
25 *       be set when the trace event is enabled.
26 */
27#define TRACE_CGROUP_PATH(type, cgrp, ...)				\
28	do {								\
29		if (trace_cgroup_##type##_enabled()) {			\
30			unsigned long flags;				\
31			spin_lock_irqsave(&trace_cgroup_path_lock,	\
32					  flags);			\
33			cgroup_path(cgrp, trace_cgroup_path,		\
34				    TRACE_CGROUP_PATH_LEN);		\
35			trace_cgroup_##type(cgrp, trace_cgroup_path,	\
36					    ##__VA_ARGS__);		\
37			spin_unlock_irqrestore(&trace_cgroup_path_lock, \
38					       flags);			\
39		}							\
40	} while (0)
41
42/*
43 * The cgroup filesystem superblock creation/mount context.
44 */
45struct cgroup_fs_context {
46	struct kernfs_fs_context kfc;
47	struct cgroup_root	*root;
48	struct cgroup_namespace	*ns;
49	unsigned int	flags;			/* CGRP_ROOT_* flags */
50
51	/* cgroup1 bits */
52	bool		cpuset_clone_children;
53	bool		none;			/* User explicitly requested empty subsystem */
54	bool		all_ss;			/* Seen 'all' option */
55	u16		subsys_mask;		/* Selected subsystems */
56	char		*name;			/* Hierarchy name */
57	char		*release_agent;		/* Path for release notifications */
58};
59
60static inline struct cgroup_fs_context *cgroup_fc2context(struct fs_context *fc)
61{
62	struct kernfs_fs_context *kfc = fc->fs_private;
63
64	return container_of(kfc, struct cgroup_fs_context, kfc);
65}
66
67struct cgroup_pidlist;
68
69struct cgroup_file_ctx {
70	struct cgroup_namespace	*ns;
71
72	struct {
73		void			*trigger;
74	} psi;
75
76	struct {
77		bool			started;
78		struct css_task_iter	iter;
79	} procs;
80
81	struct {
82		struct cgroup_pidlist	*pidlist;
83	} procs1;
84};
85
86/*
87 * A cgroup can be associated with multiple css_sets as different tasks may
88 * belong to different cgroups on different hierarchies.  In the other
89 * direction, a css_set is naturally associated with multiple cgroups.
90 * This M:N relationship is represented by the following link structure
91 * which exists for each association and allows traversing the associations
92 * from both sides.
93 */
94struct cgrp_cset_link {
95	/* the cgroup and css_set this link associates */
96	struct cgroup		*cgrp;
97	struct css_set		*cset;
98
99	/* list of cgrp_cset_links anchored at cgrp->cset_links */
100	struct list_head	cset_link;
101
102	/* list of cgrp_cset_links anchored at css_set->cgrp_links */
103	struct list_head	cgrp_link;
104};
105
106/* used to track tasks and csets during migration */
107struct cgroup_taskset {
108	/* the src and dst cset list running through cset->mg_node */
109	struct list_head	src_csets;
110	struct list_head	dst_csets;
111
112	/* the number of tasks in the set */
113	int			nr_tasks;
114
115	/* the subsys currently being processed */
116	int			ssid;
117
118	/*
119	 * Fields for cgroup_taskset_*() iteration.
120	 *
121	 * Before migration is committed, the target migration tasks are on
122	 * ->mg_tasks of the csets on ->src_csets.  After, on ->mg_tasks of
123	 * the csets on ->dst_csets.  ->csets point to either ->src_csets
124	 * or ->dst_csets depending on whether migration is committed.
125	 *
126	 * ->cur_csets and ->cur_task point to the current task position
127	 * during iteration.
128	 */
129	struct list_head	*csets;
130	struct css_set		*cur_cset;
131	struct task_struct	*cur_task;
132};
133
134/* migration context also tracks preloading */
135struct cgroup_mgctx {
136	/*
137	 * Preloaded source and destination csets.  Used to guarantee
138	 * atomic success or failure on actual migration.
139	 */
140	struct list_head	preloaded_src_csets;
141	struct list_head	preloaded_dst_csets;
142
143	/* tasks and csets to migrate */
144	struct cgroup_taskset	tset;
145
146	/* subsystems affected by migration */
147	u16			ss_mask;
148};
149
150#define CGROUP_TASKSET_INIT(tset)						\
151{										\
152	.src_csets		= LIST_HEAD_INIT(tset.src_csets),		\
153	.dst_csets		= LIST_HEAD_INIT(tset.dst_csets),		\
154	.csets			= &tset.src_csets,				\
155}
156
157#define CGROUP_MGCTX_INIT(name)							\
158{										\
159	LIST_HEAD_INIT(name.preloaded_src_csets),				\
160	LIST_HEAD_INIT(name.preloaded_dst_csets),				\
161	CGROUP_TASKSET_INIT(name.tset),						\
162}
163
164#define DEFINE_CGROUP_MGCTX(name)						\
165	struct cgroup_mgctx name = CGROUP_MGCTX_INIT(name)
166
167extern struct cgroup_subsys *cgroup_subsys[];
168extern struct list_head cgroup_roots;
169
170/* iterate across the hierarchies */
171#define for_each_root(root)						\
172	list_for_each_entry_rcu((root), &cgroup_roots, root_list,	\
173				lockdep_is_held(&cgroup_mutex))
174
175/**
176 * for_each_subsys - iterate all enabled cgroup subsystems
177 * @ss: the iteration cursor
178 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
179 */
180#define for_each_subsys(ss, ssid)					\
181	for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT &&		\
182	     (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
183
184static inline bool cgroup_is_dead(const struct cgroup *cgrp)
185{
186	return !(cgrp->self.flags & CSS_ONLINE);
187}
188
189static inline bool notify_on_release(const struct cgroup *cgrp)
190{
191	return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
192}
193
194void put_css_set_locked(struct css_set *cset);
195
196static inline void put_css_set(struct css_set *cset)
197{
198	unsigned long flags;
199
200	/*
201	 * Ensure that the refcount doesn't hit zero while any readers
202	 * can see it. Similar to atomic_dec_and_lock(), but for an
203	 * rwlock
204	 */
205	if (refcount_dec_not_one(&cset->refcount))
206		return;
207
208	spin_lock_irqsave(&css_set_lock, flags);
209	put_css_set_locked(cset);
210	spin_unlock_irqrestore(&css_set_lock, flags);
211}
212
213/*
214 * refcounted get/put for css_set objects
215 */
216static inline void get_css_set(struct css_set *cset)
217{
218	refcount_inc(&cset->refcount);
219}
220
221bool cgroup_ssid_enabled(int ssid);
222bool cgroup_on_dfl(const struct cgroup *cgrp);
223
224struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root);
225struct cgroup *task_cgroup_from_root(struct task_struct *task,
226				     struct cgroup_root *root);
227struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline);
228void cgroup_kn_unlock(struct kernfs_node *kn);
229int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
230			  struct cgroup_namespace *ns);
231
232void cgroup_favor_dynmods(struct cgroup_root *root, bool favor);
233void cgroup_free_root(struct cgroup_root *root);
234void init_cgroup_root(struct cgroup_fs_context *ctx);
235int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask);
236int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask);
237int cgroup_do_get_tree(struct fs_context *fc);
238
239int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp);
240void cgroup_migrate_finish(struct cgroup_mgctx *mgctx);
241void cgroup_migrate_add_src(struct css_set *src_cset, struct cgroup *dst_cgrp,
242			    struct cgroup_mgctx *mgctx);
243int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx);
244int cgroup_migrate(struct task_struct *leader, bool threadgroup,
245		   struct cgroup_mgctx *mgctx);
246
247int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
248		       bool threadgroup);
249void cgroup_attach_lock(bool lock_threadgroup);
250void cgroup_attach_unlock(bool lock_threadgroup);
251struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup,
252					     bool *locked)
253	__acquires(&cgroup_threadgroup_rwsem);
254void cgroup_procs_write_finish(struct task_struct *task, bool locked)
255	__releases(&cgroup_threadgroup_rwsem);
256
257void cgroup_lock_and_drain_offline(struct cgroup *cgrp);
258
259int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode);
260int cgroup_rmdir(struct kernfs_node *kn);
261int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
262		     struct kernfs_root *kf_root);
263
264int __cgroup_task_count(const struct cgroup *cgrp);
265int cgroup_task_count(const struct cgroup *cgrp);
266
267/*
268 * rstat.c
269 */
270int cgroup_rstat_init(struct cgroup *cgrp);
271void cgroup_rstat_exit(struct cgroup *cgrp);
272void cgroup_rstat_boot(void);
273void cgroup_base_stat_cputime_show(struct seq_file *seq);
274
275/*
276 * namespace.c
277 */
278extern const struct proc_ns_operations cgroupns_operations;
279
280/*
281 * cgroup-v1.c
282 */
283extern struct cftype cgroup1_base_files[];
284extern struct kernfs_syscall_ops cgroup1_kf_syscall_ops;
285extern const struct fs_parameter_spec cgroup1_fs_parameters[];
286
287int proc_cgroupstats_show(struct seq_file *m, void *v);
288bool cgroup1_ssid_disabled(int ssid);
289void cgroup1_pidlist_destroy_all(struct cgroup *cgrp);
290void cgroup1_release_agent(struct work_struct *work);
291void cgroup1_check_for_release(struct cgroup *cgrp);
292int cgroup1_parse_param(struct fs_context *fc, struct fs_parameter *param);
293int cgroup1_get_tree(struct fs_context *fc);
294int cgroup1_reconfigure(struct fs_context *ctx);
295
296#endif /* __CGROUP_INTERNAL_H */
297