xref: /linux/kernel/cgroup/cgroup-internal.h (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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
13 extern spinlock_t trace_cgroup_path_lock;
14 extern char trace_cgroup_path[TRACE_CGROUP_PATH_LEN];
15 extern 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  */
45 struct 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 
60 static 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 
67 struct cgroup_pidlist;
68 
69 struct 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 	struct cgroup_of_peak peak;
86 };
87 
88 /*
89  * A cgroup can be associated with multiple css_sets as different tasks may
90  * belong to different cgroups on different hierarchies.  In the other
91  * direction, a css_set is naturally associated with multiple cgroups.
92  * This M:N relationship is represented by the following link structure
93  * which exists for each association and allows traversing the associations
94  * from both sides.
95  */
96 struct cgrp_cset_link {
97 	/* the cgroup and css_set this link associates */
98 	struct cgroup		*cgrp;
99 	struct css_set		*cset;
100 
101 	/* list of cgrp_cset_links anchored at cgrp->cset_links */
102 	struct list_head	cset_link;
103 
104 	/* list of cgrp_cset_links anchored at css_set->cgrp_links */
105 	struct list_head	cgrp_link;
106 };
107 
108 /* used to track tasks and csets during migration */
109 struct cgroup_taskset {
110 	/* the src and dst cset list running through cset->mg_node */
111 	struct list_head	src_csets;
112 	struct list_head	dst_csets;
113 
114 	/* the number of tasks in the set */
115 	int			nr_tasks;
116 
117 	/* the subsys currently being processed */
118 	int			ssid;
119 
120 	/*
121 	 * Fields for cgroup_taskset_*() iteration.
122 	 *
123 	 * Before migration is committed, the target migration tasks are on
124 	 * ->mg_tasks of the csets on ->src_csets.  After, on ->mg_tasks of
125 	 * the csets on ->dst_csets.  ->csets point to either ->src_csets
126 	 * or ->dst_csets depending on whether migration is committed.
127 	 *
128 	 * ->cur_csets and ->cur_task point to the current task position
129 	 * during iteration.
130 	 */
131 	struct list_head	*csets;
132 	struct css_set		*cur_cset;
133 	struct task_struct	*cur_task;
134 };
135 
136 /* migration context also tracks preloading */
137 struct cgroup_mgctx {
138 	/*
139 	 * Preloaded source and destination csets.  Used to guarantee
140 	 * atomic success or failure on actual migration.
141 	 */
142 	struct list_head	preloaded_src_csets;
143 	struct list_head	preloaded_dst_csets;
144 
145 	/* tasks and csets to migrate */
146 	struct cgroup_taskset	tset;
147 
148 	/* subsystems affected by migration */
149 	u16			ss_mask;
150 };
151 
152 #define CGROUP_TASKSET_INIT(tset)						\
153 {										\
154 	.src_csets		= LIST_HEAD_INIT(tset.src_csets),		\
155 	.dst_csets		= LIST_HEAD_INIT(tset.dst_csets),		\
156 	.csets			= &tset.src_csets,				\
157 }
158 
159 #define CGROUP_MGCTX_INIT(name)							\
160 {										\
161 	LIST_HEAD_INIT(name.preloaded_src_csets),				\
162 	LIST_HEAD_INIT(name.preloaded_dst_csets),				\
163 	CGROUP_TASKSET_INIT(name.tset),						\
164 }
165 
166 #define DEFINE_CGROUP_MGCTX(name)						\
167 	struct cgroup_mgctx name = CGROUP_MGCTX_INIT(name)
168 
169 extern struct cgroup_subsys *cgroup_subsys[];
170 extern struct list_head cgroup_roots;
171 
172 /* iterate across the hierarchies */
173 #define for_each_root(root)						\
174 	list_for_each_entry_rcu((root), &cgroup_roots, root_list,	\
175 				lockdep_is_held(&cgroup_mutex))
176 
177 /**
178  * for_each_subsys - iterate all enabled cgroup subsystems
179  * @ss: the iteration cursor
180  * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
181  */
182 #define for_each_subsys(ss, ssid)					\
183 	for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT &&		\
184 	     (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
185 
186 static inline bool cgroup_is_dead(const struct cgroup *cgrp)
187 {
188 	return !(cgrp->self.flags & CSS_ONLINE);
189 }
190 
191 static inline bool notify_on_release(const struct cgroup *cgrp)
192 {
193 	return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
194 }
195 
196 void put_css_set_locked(struct css_set *cset);
197 
198 static inline void put_css_set(struct css_set *cset)
199 {
200 	unsigned long flags;
201 
202 	/*
203 	 * Ensure that the refcount doesn't hit zero while any readers
204 	 * can see it. Similar to atomic_dec_and_lock(), but for an
205 	 * rwlock
206 	 */
207 	if (refcount_dec_not_one(&cset->refcount))
208 		return;
209 
210 	spin_lock_irqsave(&css_set_lock, flags);
211 	put_css_set_locked(cset);
212 	spin_unlock_irqrestore(&css_set_lock, flags);
213 }
214 
215 /*
216  * refcounted get/put for css_set objects
217  */
218 static inline void get_css_set(struct css_set *cset)
219 {
220 	refcount_inc(&cset->refcount);
221 }
222 
223 bool cgroup_ssid_enabled(int ssid);
224 bool cgroup_on_dfl(const struct cgroup *cgrp);
225 
226 struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root);
227 struct cgroup *task_cgroup_from_root(struct task_struct *task,
228 				     struct cgroup_root *root);
229 struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline);
230 void cgroup_kn_unlock(struct kernfs_node *kn);
231 int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
232 			  struct cgroup_namespace *ns);
233 
234 void cgroup_favor_dynmods(struct cgroup_root *root, bool favor);
235 void cgroup_free_root(struct cgroup_root *root);
236 void init_cgroup_root(struct cgroup_fs_context *ctx);
237 int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask);
238 int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask);
239 int cgroup_do_get_tree(struct fs_context *fc);
240 
241 int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp);
242 void cgroup_migrate_finish(struct cgroup_mgctx *mgctx);
243 void cgroup_migrate_add_src(struct css_set *src_cset, struct cgroup *dst_cgrp,
244 			    struct cgroup_mgctx *mgctx);
245 int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx);
246 int cgroup_migrate(struct task_struct *leader, bool threadgroup,
247 		   struct cgroup_mgctx *mgctx);
248 
249 int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
250 		       bool threadgroup);
251 void cgroup_attach_lock(bool lock_threadgroup);
252 void cgroup_attach_unlock(bool lock_threadgroup);
253 struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup,
254 					     bool *locked)
255 	__acquires(&cgroup_threadgroup_rwsem);
256 void cgroup_procs_write_finish(struct task_struct *task, bool locked)
257 	__releases(&cgroup_threadgroup_rwsem);
258 
259 void cgroup_lock_and_drain_offline(struct cgroup *cgrp);
260 
261 int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode);
262 int cgroup_rmdir(struct kernfs_node *kn);
263 int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
264 		     struct kernfs_root *kf_root);
265 
266 int __cgroup_task_count(const struct cgroup *cgrp);
267 int cgroup_task_count(const struct cgroup *cgrp);
268 
269 /*
270  * rstat.c
271  */
272 int cgroup_rstat_init(struct cgroup *cgrp);
273 void cgroup_rstat_exit(struct cgroup *cgrp);
274 void cgroup_rstat_boot(void);
275 void cgroup_base_stat_cputime_show(struct seq_file *seq);
276 
277 /*
278  * namespace.c
279  */
280 extern const struct proc_ns_operations cgroupns_operations;
281 
282 /*
283  * cgroup-v1.c
284  */
285 extern struct cftype cgroup1_base_files[];
286 extern struct kernfs_syscall_ops cgroup1_kf_syscall_ops;
287 extern const struct fs_parameter_spec cgroup1_fs_parameters[];
288 
289 int proc_cgroupstats_show(struct seq_file *m, void *v);
290 bool cgroup1_ssid_disabled(int ssid);
291 void cgroup1_pidlist_destroy_all(struct cgroup *cgrp);
292 void cgroup1_release_agent(struct work_struct *work);
293 void cgroup1_check_for_release(struct cgroup *cgrp);
294 int cgroup1_parse_param(struct fs_context *fc, struct fs_parameter *param);
295 int cgroup1_get_tree(struct fs_context *fc);
296 int cgroup1_reconfigure(struct fs_context *ctx);
297 
298 #endif /* __CGROUP_INTERNAL_H */
299