xref: /linux-6.15/include/linux/sched/task.h (revision 2a52ca7c)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_SCHED_TASK_H
3 #define _LINUX_SCHED_TASK_H
4 
5 /*
6  * Interface between the scheduler and various task lifetime (fork()/exit())
7  * functionality:
8  */
9 
10 #include <linux/rcupdate.h>
11 #include <linux/refcount.h>
12 #include <linux/sched.h>
13 #include <linux/uaccess.h>
14 
15 struct task_struct;
16 struct rusage;
17 union thread_union;
18 struct css_set;
19 
20 /* All the bits taken by the old clone syscall. */
21 #define CLONE_LEGACY_FLAGS 0xffffffffULL
22 
23 struct kernel_clone_args {
24 	u64 flags;
25 	int __user *pidfd;
26 	int __user *child_tid;
27 	int __user *parent_tid;
28 	const char *name;
29 	int exit_signal;
30 	u32 kthread:1;
31 	u32 io_thread:1;
32 	u32 user_worker:1;
33 	u32 no_files:1;
34 	unsigned long stack;
35 	unsigned long stack_size;
36 	unsigned long tls;
37 	pid_t *set_tid;
38 	/* Number of elements in *set_tid */
39 	size_t set_tid_size;
40 	int cgroup;
41 	int idle;
42 	int (*fn)(void *);
43 	void *fn_arg;
44 	struct cgroup *cgrp;
45 	struct css_set *cset;
46 };
47 
48 /*
49  * This serializes "schedule()" and also protects
50  * the run-queue from deletions/modifications (but
51  * _adding_ to the beginning of the run-queue has
52  * a separate lock).
53  */
54 extern rwlock_t tasklist_lock;
55 extern spinlock_t mmlist_lock;
56 
57 extern union thread_union init_thread_union;
58 extern struct task_struct init_task;
59 
60 extern int lockdep_tasklist_lock_is_held(void);
61 
62 extern asmlinkage void schedule_tail(struct task_struct *prev);
63 extern void init_idle(struct task_struct *idle, int cpu);
64 
65 extern int sched_fork(unsigned long clone_flags, struct task_struct *p);
66 extern int sched_cgroup_fork(struct task_struct *p, struct kernel_clone_args *kargs);
67 extern void sched_cancel_fork(struct task_struct *p);
68 extern void sched_post_fork(struct task_struct *p);
69 extern void sched_dead(struct task_struct *p);
70 
71 void __noreturn do_task_dead(void);
72 void __noreturn make_task_dead(int signr);
73 
74 extern void mm_cache_init(void);
75 extern void proc_caches_init(void);
76 
77 extern void fork_init(void);
78 
79 extern void release_task(struct task_struct * p);
80 
81 extern int copy_thread(struct task_struct *, const struct kernel_clone_args *);
82 
83 extern void flush_thread(void);
84 
85 #ifdef CONFIG_HAVE_EXIT_THREAD
86 extern void exit_thread(struct task_struct *tsk);
87 #else
88 static inline void exit_thread(struct task_struct *tsk)
89 {
90 }
91 #endif
92 extern __noreturn void do_group_exit(int);
93 
94 extern void exit_files(struct task_struct *);
95 extern void exit_itimers(struct task_struct *);
96 
97 extern pid_t kernel_clone(struct kernel_clone_args *kargs);
98 struct task_struct *copy_process(struct pid *pid, int trace, int node,
99 				 struct kernel_clone_args *args);
100 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node);
101 struct task_struct *fork_idle(int);
102 extern pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
103 			    unsigned long flags);
104 extern pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags);
105 extern long kernel_wait4(pid_t, int __user *, int, struct rusage *);
106 int kernel_wait(pid_t pid, int *stat);
107 
108 extern void free_task(struct task_struct *tsk);
109 
110 /* sched_exec is called by processes performing an exec */
111 #ifdef CONFIG_SMP
112 extern void sched_exec(void);
113 #else
114 #define sched_exec()   {}
115 #endif
116 
117 static inline struct task_struct *get_task_struct(struct task_struct *t)
118 {
119 	refcount_inc(&t->usage);
120 	return t;
121 }
122 
123 extern void __put_task_struct(struct task_struct *t);
124 extern void __put_task_struct_rcu_cb(struct rcu_head *rhp);
125 
126 static inline void put_task_struct(struct task_struct *t)
127 {
128 	if (!refcount_dec_and_test(&t->usage))
129 		return;
130 
131 	/*
132 	 * In !RT, it is always safe to call __put_task_struct().
133 	 * Under RT, we can only call it in preemptible context.
134 	 */
135 	if (!IS_ENABLED(CONFIG_PREEMPT_RT) || preemptible()) {
136 		static DEFINE_WAIT_OVERRIDE_MAP(put_task_map, LD_WAIT_SLEEP);
137 
138 		lock_map_acquire_try(&put_task_map);
139 		__put_task_struct(t);
140 		lock_map_release(&put_task_map);
141 		return;
142 	}
143 
144 	/*
145 	 * under PREEMPT_RT, we can't call put_task_struct
146 	 * in atomic context because it will indirectly
147 	 * acquire sleeping locks.
148 	 *
149 	 * call_rcu() will schedule delayed_put_task_struct_rcu()
150 	 * to be called in process context.
151 	 *
152 	 * __put_task_struct() is called when
153 	 * refcount_dec_and_test(&t->usage) succeeds.
154 	 *
155 	 * This means that it can't "conflict" with
156 	 * put_task_struct_rcu_user() which abuses ->rcu the same
157 	 * way; rcu_users has a reference so task->usage can't be
158 	 * zero after rcu_users 1 -> 0 transition.
159 	 *
160 	 * delayed_free_task() also uses ->rcu, but it is only called
161 	 * when it fails to fork a process. Therefore, there is no
162 	 * way it can conflict with put_task_struct().
163 	 */
164 	call_rcu(&t->rcu, __put_task_struct_rcu_cb);
165 }
166 
167 DEFINE_FREE(put_task, struct task_struct *, if (_T) put_task_struct(_T))
168 
169 static inline void put_task_struct_many(struct task_struct *t, int nr)
170 {
171 	if (refcount_sub_and_test(nr, &t->usage))
172 		__put_task_struct(t);
173 }
174 
175 void put_task_struct_rcu_user(struct task_struct *task);
176 
177 /* Free all architecture-specific resources held by a thread. */
178 void release_thread(struct task_struct *dead_task);
179 
180 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
181 extern int arch_task_struct_size __read_mostly;
182 #else
183 # define arch_task_struct_size (sizeof(struct task_struct))
184 #endif
185 
186 #ifndef CONFIG_HAVE_ARCH_THREAD_STRUCT_WHITELIST
187 /*
188  * If an architecture has not declared a thread_struct whitelist we
189  * must assume something there may need to be copied to userspace.
190  */
191 static inline void arch_thread_struct_whitelist(unsigned long *offset,
192 						unsigned long *size)
193 {
194 	*offset = 0;
195 	/* Handle dynamically sized thread_struct. */
196 	*size = arch_task_struct_size - offsetof(struct task_struct, thread);
197 }
198 #endif
199 
200 #ifdef CONFIG_VMAP_STACK
201 static inline struct vm_struct *task_stack_vm_area(const struct task_struct *t)
202 {
203 	return t->stack_vm_area;
204 }
205 #else
206 static inline struct vm_struct *task_stack_vm_area(const struct task_struct *t)
207 {
208 	return NULL;
209 }
210 #endif
211 
212 /*
213  * Protects ->fs, ->files, ->mm, ->group_info, ->comm, keyring
214  * subscriptions and synchronises with wait4().  Also used in procfs.  Also
215  * pins the final release of task.io_context.  Also protects ->cpuset and
216  * ->cgroup.subsys[]. And ->vfork_done. And ->sysvshm.shm_clist.
217  *
218  * Nests both inside and outside of read_lock(&tasklist_lock).
219  * It must not be nested with write_lock_irq(&tasklist_lock),
220  * neither inside nor outside.
221  */
222 static inline void task_lock(struct task_struct *p)
223 {
224 	spin_lock(&p->alloc_lock);
225 }
226 
227 static inline void task_unlock(struct task_struct *p)
228 {
229 	spin_unlock(&p->alloc_lock);
230 }
231 
232 DEFINE_GUARD(task_lock, struct task_struct *, task_lock(_T), task_unlock(_T))
233 
234 #endif /* _LINUX_SCHED_TASK_H */
235