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 
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  */
94 struct 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 */
107 struct 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 */
135 struct 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 
167 extern spinlock_t css_set_lock;
168 extern struct cgroup_subsys *cgroup_subsys[];
169 extern struct list_head cgroup_roots;
170 extern struct file_system_type cgroup_fs_type;
171 
172 /* iterate across the hierarchies */
173 #define for_each_root(root)						\
174 	list_for_each_entry((root), &cgroup_roots, root_list)
175 
176 /**
177  * for_each_subsys - iterate all enabled cgroup subsystems
178  * @ss: the iteration cursor
179  * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
180  */
181 #define for_each_subsys(ss, ssid)					\
182 	for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT &&		\
183 	     (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
184 
185 static inline bool cgroup_is_dead(const struct cgroup *cgrp)
186 {
187 	return !(cgrp->self.flags & CSS_ONLINE);
188 }
189 
190 static inline bool notify_on_release(const struct cgroup *cgrp)
191 {
192 	return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
193 }
194 
195 void put_css_set_locked(struct css_set *cset);
196 
197 static inline void put_css_set(struct css_set *cset)
198 {
199 	unsigned long flags;
200 
201 	/*
202 	 * Ensure that the refcount doesn't hit zero while any readers
203 	 * can see it. Similar to atomic_dec_and_lock(), but for an
204 	 * rwlock
205 	 */
206 	if (refcount_dec_not_one(&cset->refcount))
207 		return;
208 
209 	spin_lock_irqsave(&css_set_lock, flags);
210 	put_css_set_locked(cset);
211 	spin_unlock_irqrestore(&css_set_lock, flags);
212 }
213 
214 /*
215  * refcounted get/put for css_set objects
216  */
217 static inline void get_css_set(struct css_set *cset)
218 {
219 	refcount_inc(&cset->refcount);
220 }
221 
222 bool cgroup_ssid_enabled(int ssid);
223 bool cgroup_on_dfl(const struct cgroup *cgrp);
224 bool cgroup_is_thread_root(struct cgroup *cgrp);
225 bool cgroup_is_threaded(struct cgroup *cgrp);
226 
227 struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root);
228 struct cgroup *task_cgroup_from_root(struct task_struct *task,
229 				     struct cgroup_root *root);
230 struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline);
231 void cgroup_kn_unlock(struct kernfs_node *kn);
232 int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
233 			  struct cgroup_namespace *ns);
234 
235 void cgroup_favor_dynmods(struct cgroup_root *root, bool favor);
236 void cgroup_free_root(struct cgroup_root *root);
237 void init_cgroup_root(struct cgroup_fs_context *ctx);
238 int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask);
239 int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask);
240 int cgroup_do_get_tree(struct fs_context *fc);
241 
242 int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp);
243 void cgroup_migrate_finish(struct cgroup_mgctx *mgctx);
244 void cgroup_migrate_add_src(struct css_set *src_cset, struct cgroup *dst_cgrp,
245 			    struct cgroup_mgctx *mgctx);
246 int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx);
247 int cgroup_migrate(struct task_struct *leader, bool threadgroup,
248 		   struct cgroup_mgctx *mgctx);
249 
250 int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
251 		       bool threadgroup);
252 void cgroup_attach_lock(bool lock_threadgroup);
253 void cgroup_attach_unlock(bool lock_threadgroup);
254 struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup,
255 					     bool *locked)
256 	__acquires(&cgroup_threadgroup_rwsem);
257 void cgroup_procs_write_finish(struct task_struct *task, bool locked)
258 	__releases(&cgroup_threadgroup_rwsem);
259 
260 void cgroup_lock_and_drain_offline(struct cgroup *cgrp);
261 
262 int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode);
263 int cgroup_rmdir(struct kernfs_node *kn);
264 int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
265 		     struct kernfs_root *kf_root);
266 
267 int __cgroup_task_count(const struct cgroup *cgrp);
268 int cgroup_task_count(const struct cgroup *cgrp);
269 
270 /*
271  * rstat.c
272  */
273 int cgroup_rstat_init(struct cgroup *cgrp);
274 void cgroup_rstat_exit(struct cgroup *cgrp);
275 void cgroup_rstat_boot(void);
276 void cgroup_base_stat_cputime_show(struct seq_file *seq);
277 
278 /*
279  * namespace.c
280  */
281 extern const struct proc_ns_operations cgroupns_operations;
282 
283 /*
284  * cgroup-v1.c
285  */
286 extern struct cftype cgroup1_base_files[];
287 extern struct kernfs_syscall_ops cgroup1_kf_syscall_ops;
288 extern const struct fs_parameter_spec cgroup1_fs_parameters[];
289 
290 int proc_cgroupstats_show(struct seq_file *m, void *v);
291 bool cgroup1_ssid_disabled(int ssid);
292 void cgroup1_pidlist_destroy_all(struct cgroup *cgrp);
293 void cgroup1_release_agent(struct work_struct *work);
294 void cgroup1_check_for_release(struct cgroup *cgrp);
295 int cgroup1_parse_param(struct fs_context *fc, struct fs_parameter *param);
296 int cgroup1_get_tree(struct fs_context *fc);
297 int cgroup1_reconfigure(struct fs_context *ctx);
298 
299 #endif /* __CGROUP_INTERNAL_H */
300