xref: /linux/kernel/cgroup/cgroup-internal.h (revision e9f0878c4b2004ac19581274c1ae4c61ae3ca70e)
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 
11 #define TRACE_CGROUP_PATH_LEN 1024
12 extern spinlock_t trace_cgroup_path_lock;
13 extern char trace_cgroup_path[TRACE_CGROUP_PATH_LEN];
14 
15 /*
16  * cgroup_path() takes a spin lock. It is good practice not to take
17  * spin locks within trace point handlers, as they are mostly hidden
18  * from normal view. As cgroup_path() can take the kernfs_rename_lock
19  * spin lock, it is best to not call that function from the trace event
20  * handler.
21  *
22  * Note: trace_cgroup_##type##_enabled() is a static branch that will only
23  *       be set when the trace event is enabled.
24  */
25 #define TRACE_CGROUP_PATH(type, cgrp, ...)				\
26 	do {								\
27 		if (trace_cgroup_##type##_enabled()) {			\
28 			spin_lock(&trace_cgroup_path_lock);		\
29 			cgroup_path(cgrp, trace_cgroup_path,		\
30 				    TRACE_CGROUP_PATH_LEN);		\
31 			trace_cgroup_##type(cgrp, trace_cgroup_path,	\
32 					    ##__VA_ARGS__);		\
33 			spin_unlock(&trace_cgroup_path_lock);		\
34 		}							\
35 	} while (0)
36 
37 /*
38  * A cgroup can be associated with multiple css_sets as different tasks may
39  * belong to different cgroups on different hierarchies.  In the other
40  * direction, a css_set is naturally associated with multiple cgroups.
41  * This M:N relationship is represented by the following link structure
42  * which exists for each association and allows traversing the associations
43  * from both sides.
44  */
45 struct cgrp_cset_link {
46 	/* the cgroup and css_set this link associates */
47 	struct cgroup		*cgrp;
48 	struct css_set		*cset;
49 
50 	/* list of cgrp_cset_links anchored at cgrp->cset_links */
51 	struct list_head	cset_link;
52 
53 	/* list of cgrp_cset_links anchored at css_set->cgrp_links */
54 	struct list_head	cgrp_link;
55 };
56 
57 /* used to track tasks and csets during migration */
58 struct cgroup_taskset {
59 	/* the src and dst cset list running through cset->mg_node */
60 	struct list_head	src_csets;
61 	struct list_head	dst_csets;
62 
63 	/* the number of tasks in the set */
64 	int			nr_tasks;
65 
66 	/* the subsys currently being processed */
67 	int			ssid;
68 
69 	/*
70 	 * Fields for cgroup_taskset_*() iteration.
71 	 *
72 	 * Before migration is committed, the target migration tasks are on
73 	 * ->mg_tasks of the csets on ->src_csets.  After, on ->mg_tasks of
74 	 * the csets on ->dst_csets.  ->csets point to either ->src_csets
75 	 * or ->dst_csets depending on whether migration is committed.
76 	 *
77 	 * ->cur_csets and ->cur_task point to the current task position
78 	 * during iteration.
79 	 */
80 	struct list_head	*csets;
81 	struct css_set		*cur_cset;
82 	struct task_struct	*cur_task;
83 };
84 
85 /* migration context also tracks preloading */
86 struct cgroup_mgctx {
87 	/*
88 	 * Preloaded source and destination csets.  Used to guarantee
89 	 * atomic success or failure on actual migration.
90 	 */
91 	struct list_head	preloaded_src_csets;
92 	struct list_head	preloaded_dst_csets;
93 
94 	/* tasks and csets to migrate */
95 	struct cgroup_taskset	tset;
96 
97 	/* subsystems affected by migration */
98 	u16			ss_mask;
99 };
100 
101 #define CGROUP_TASKSET_INIT(tset)						\
102 {										\
103 	.src_csets		= LIST_HEAD_INIT(tset.src_csets),		\
104 	.dst_csets		= LIST_HEAD_INIT(tset.dst_csets),		\
105 	.csets			= &tset.src_csets,				\
106 }
107 
108 #define CGROUP_MGCTX_INIT(name)							\
109 {										\
110 	LIST_HEAD_INIT(name.preloaded_src_csets),				\
111 	LIST_HEAD_INIT(name.preloaded_dst_csets),				\
112 	CGROUP_TASKSET_INIT(name.tset),						\
113 }
114 
115 #define DEFINE_CGROUP_MGCTX(name)						\
116 	struct cgroup_mgctx name = CGROUP_MGCTX_INIT(name)
117 
118 struct cgroup_sb_opts {
119 	u16 subsys_mask;
120 	unsigned int flags;
121 	char *release_agent;
122 	bool cpuset_clone_children;
123 	char *name;
124 	/* User explicitly requested empty subsystem */
125 	bool none;
126 };
127 
128 extern struct mutex cgroup_mutex;
129 extern spinlock_t css_set_lock;
130 extern struct cgroup_subsys *cgroup_subsys[];
131 extern struct list_head cgroup_roots;
132 extern struct file_system_type cgroup_fs_type;
133 
134 /* iterate across the hierarchies */
135 #define for_each_root(root)						\
136 	list_for_each_entry((root), &cgroup_roots, root_list)
137 
138 /**
139  * for_each_subsys - iterate all enabled cgroup subsystems
140  * @ss: the iteration cursor
141  * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
142  */
143 #define for_each_subsys(ss, ssid)					\
144 	for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT &&		\
145 	     (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
146 
147 static inline bool cgroup_is_dead(const struct cgroup *cgrp)
148 {
149 	return !(cgrp->self.flags & CSS_ONLINE);
150 }
151 
152 static inline bool notify_on_release(const struct cgroup *cgrp)
153 {
154 	return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
155 }
156 
157 void put_css_set_locked(struct css_set *cset);
158 
159 static inline void put_css_set(struct css_set *cset)
160 {
161 	unsigned long flags;
162 
163 	/*
164 	 * Ensure that the refcount doesn't hit zero while any readers
165 	 * can see it. Similar to atomic_dec_and_lock(), but for an
166 	 * rwlock
167 	 */
168 	if (refcount_dec_not_one(&cset->refcount))
169 		return;
170 
171 	spin_lock_irqsave(&css_set_lock, flags);
172 	put_css_set_locked(cset);
173 	spin_unlock_irqrestore(&css_set_lock, flags);
174 }
175 
176 /*
177  * refcounted get/put for css_set objects
178  */
179 static inline void get_css_set(struct css_set *cset)
180 {
181 	refcount_inc(&cset->refcount);
182 }
183 
184 bool cgroup_ssid_enabled(int ssid);
185 bool cgroup_on_dfl(const struct cgroup *cgrp);
186 bool cgroup_is_thread_root(struct cgroup *cgrp);
187 bool cgroup_is_threaded(struct cgroup *cgrp);
188 
189 struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root);
190 struct cgroup *task_cgroup_from_root(struct task_struct *task,
191 				     struct cgroup_root *root);
192 struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline);
193 void cgroup_kn_unlock(struct kernfs_node *kn);
194 int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
195 			  struct cgroup_namespace *ns);
196 
197 void cgroup_free_root(struct cgroup_root *root);
198 void init_cgroup_root(struct cgroup_root *root, struct cgroup_sb_opts *opts);
199 int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask, int ref_flags);
200 int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask);
201 struct dentry *cgroup_do_mount(struct file_system_type *fs_type, int flags,
202 			       struct cgroup_root *root, unsigned long magic,
203 			       struct cgroup_namespace *ns);
204 
205 int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp);
206 void cgroup_migrate_finish(struct cgroup_mgctx *mgctx);
207 void cgroup_migrate_add_src(struct css_set *src_cset, struct cgroup *dst_cgrp,
208 			    struct cgroup_mgctx *mgctx);
209 int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx);
210 int cgroup_migrate(struct task_struct *leader, bool threadgroup,
211 		   struct cgroup_mgctx *mgctx);
212 
213 int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
214 		       bool threadgroup);
215 struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup)
216 	__acquires(&cgroup_threadgroup_rwsem);
217 void cgroup_procs_write_finish(struct task_struct *task)
218 	__releases(&cgroup_threadgroup_rwsem);
219 
220 void cgroup_lock_and_drain_offline(struct cgroup *cgrp);
221 
222 int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode);
223 int cgroup_rmdir(struct kernfs_node *kn);
224 int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
225 		     struct kernfs_root *kf_root);
226 
227 int cgroup_task_count(const struct cgroup *cgrp);
228 
229 /*
230  * rstat.c
231  */
232 int cgroup_rstat_init(struct cgroup *cgrp);
233 void cgroup_rstat_exit(struct cgroup *cgrp);
234 void cgroup_rstat_boot(void);
235 void cgroup_base_stat_cputime_show(struct seq_file *seq);
236 
237 /*
238  * namespace.c
239  */
240 extern const struct proc_ns_operations cgroupns_operations;
241 
242 /*
243  * cgroup-v1.c
244  */
245 extern struct cftype cgroup1_base_files[];
246 extern struct kernfs_syscall_ops cgroup1_kf_syscall_ops;
247 
248 int proc_cgroupstats_show(struct seq_file *m, void *v);
249 bool cgroup1_ssid_disabled(int ssid);
250 void cgroup1_pidlist_destroy_all(struct cgroup *cgrp);
251 void cgroup1_release_agent(struct work_struct *work);
252 void cgroup1_check_for_release(struct cgroup *cgrp);
253 struct dentry *cgroup1_mount(struct file_system_type *fs_type, int flags,
254 			     void *data, unsigned long magic,
255 			     struct cgroup_namespace *ns);
256 
257 #endif /* __CGROUP_INTERNAL_H */
258