xref: /linux-6.15/include/linux/sched/mm.h (revision 66dfdff0)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_SCHED_MM_H
3 #define _LINUX_SCHED_MM_H
4 
5 #include <linux/kernel.h>
6 #include <linux/atomic.h>
7 #include <linux/sched.h>
8 #include <linux/mm_types.h>
9 #include <linux/gfp.h>
10 #include <linux/sync_core.h>
11 
12 /*
13  * Routines for handling mm_structs
14  */
15 extern struct mm_struct *mm_alloc(void);
16 
17 /**
18  * mmgrab() - Pin a &struct mm_struct.
19  * @mm: The &struct mm_struct to pin.
20  *
21  * Make sure that @mm will not get freed even after the owning task
22  * exits. This doesn't guarantee that the associated address space
23  * will still exist later on and mmget_not_zero() has to be used before
24  * accessing it.
25  *
26  * This is a preferred way to to pin @mm for a longer/unbounded amount
27  * of time.
28  *
29  * Use mmdrop() to release the reference acquired by mmgrab().
30  *
31  * See also <Documentation/vm/active_mm.txt> for an in-depth explanation
32  * of &mm_struct.mm_count vs &mm_struct.mm_users.
33  */
34 static inline void mmgrab(struct mm_struct *mm)
35 {
36 	atomic_inc(&mm->mm_count);
37 }
38 
39 extern void mmdrop(struct mm_struct *mm);
40 
41 /**
42  * mmget() - Pin the address space associated with a &struct mm_struct.
43  * @mm: The address space to pin.
44  *
45  * Make sure that the address space of the given &struct mm_struct doesn't
46  * go away. This does not protect against parts of the address space being
47  * modified or freed, however.
48  *
49  * Never use this function to pin this address space for an
50  * unbounded/indefinite amount of time.
51  *
52  * Use mmput() to release the reference acquired by mmget().
53  *
54  * See also <Documentation/vm/active_mm.txt> for an in-depth explanation
55  * of &mm_struct.mm_count vs &mm_struct.mm_users.
56  */
57 static inline void mmget(struct mm_struct *mm)
58 {
59 	atomic_inc(&mm->mm_users);
60 }
61 
62 static inline bool mmget_not_zero(struct mm_struct *mm)
63 {
64 	return atomic_inc_not_zero(&mm->mm_users);
65 }
66 
67 /* mmput gets rid of the mappings and all user-space */
68 extern void mmput(struct mm_struct *);
69 #ifdef CONFIG_MMU
70 /* same as above but performs the slow path from the async context. Can
71  * be called from the atomic context as well
72  */
73 void mmput_async(struct mm_struct *);
74 #endif
75 
76 /* Grab a reference to a task's mm, if it is not already going away */
77 extern struct mm_struct *get_task_mm(struct task_struct *task);
78 /*
79  * Grab a reference to a task's mm, if it is not already going away
80  * and ptrace_may_access with the mode parameter passed to it
81  * succeeds.
82  */
83 extern struct mm_struct *mm_access(struct task_struct *task, unsigned int mode);
84 /* Remove the current tasks stale references to the old mm_struct */
85 extern void mm_release(struct task_struct *, struct mm_struct *);
86 
87 #ifdef CONFIG_MEMCG
88 extern void mm_update_next_owner(struct mm_struct *mm);
89 #else
90 static inline void mm_update_next_owner(struct mm_struct *mm)
91 {
92 }
93 #endif /* CONFIG_MEMCG */
94 
95 #ifdef CONFIG_MMU
96 extern void arch_pick_mmap_layout(struct mm_struct *mm);
97 extern unsigned long
98 arch_get_unmapped_area(struct file *, unsigned long, unsigned long,
99 		       unsigned long, unsigned long);
100 extern unsigned long
101 arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
102 			  unsigned long len, unsigned long pgoff,
103 			  unsigned long flags);
104 #else
105 static inline void arch_pick_mmap_layout(struct mm_struct *mm) {}
106 #endif
107 
108 static inline bool in_vfork(struct task_struct *tsk)
109 {
110 	bool ret;
111 
112 	/*
113 	 * need RCU to access ->real_parent if CLONE_VM was used along with
114 	 * CLONE_PARENT.
115 	 *
116 	 * We check real_parent->mm == tsk->mm because CLONE_VFORK does not
117 	 * imply CLONE_VM
118 	 *
119 	 * CLONE_VFORK can be used with CLONE_PARENT/CLONE_THREAD and thus
120 	 * ->real_parent is not necessarily the task doing vfork(), so in
121 	 * theory we can't rely on task_lock() if we want to dereference it.
122 	 *
123 	 * And in this case we can't trust the real_parent->mm == tsk->mm
124 	 * check, it can be false negative. But we do not care, if init or
125 	 * another oom-unkillable task does this it should blame itself.
126 	 */
127 	rcu_read_lock();
128 	ret = tsk->vfork_done && tsk->real_parent->mm == tsk->mm;
129 	rcu_read_unlock();
130 
131 	return ret;
132 }
133 
134 /*
135  * Applies per-task gfp context to the given allocation flags.
136  * PF_MEMALLOC_NOIO implies GFP_NOIO
137  * PF_MEMALLOC_NOFS implies GFP_NOFS
138  */
139 static inline gfp_t current_gfp_context(gfp_t flags)
140 {
141 	/*
142 	 * NOIO implies both NOIO and NOFS and it is a weaker context
143 	 * so always make sure it makes precendence
144 	 */
145 	if (unlikely(current->flags & PF_MEMALLOC_NOIO))
146 		flags &= ~(__GFP_IO | __GFP_FS);
147 	else if (unlikely(current->flags & PF_MEMALLOC_NOFS))
148 		flags &= ~__GFP_FS;
149 	return flags;
150 }
151 
152 #ifdef CONFIG_LOCKDEP
153 extern void fs_reclaim_acquire(gfp_t gfp_mask);
154 extern void fs_reclaim_release(gfp_t gfp_mask);
155 #else
156 static inline void fs_reclaim_acquire(gfp_t gfp_mask) { }
157 static inline void fs_reclaim_release(gfp_t gfp_mask) { }
158 #endif
159 
160 static inline unsigned int memalloc_noio_save(void)
161 {
162 	unsigned int flags = current->flags & PF_MEMALLOC_NOIO;
163 	current->flags |= PF_MEMALLOC_NOIO;
164 	return flags;
165 }
166 
167 static inline void memalloc_noio_restore(unsigned int flags)
168 {
169 	current->flags = (current->flags & ~PF_MEMALLOC_NOIO) | flags;
170 }
171 
172 static inline unsigned int memalloc_nofs_save(void)
173 {
174 	unsigned int flags = current->flags & PF_MEMALLOC_NOFS;
175 	current->flags |= PF_MEMALLOC_NOFS;
176 	return flags;
177 }
178 
179 static inline void memalloc_nofs_restore(unsigned int flags)
180 {
181 	current->flags = (current->flags & ~PF_MEMALLOC_NOFS) | flags;
182 }
183 
184 static inline unsigned int memalloc_noreclaim_save(void)
185 {
186 	unsigned int flags = current->flags & PF_MEMALLOC;
187 	current->flags |= PF_MEMALLOC;
188 	return flags;
189 }
190 
191 static inline void memalloc_noreclaim_restore(unsigned int flags)
192 {
193 	current->flags = (current->flags & ~PF_MEMALLOC) | flags;
194 }
195 
196 #ifdef CONFIG_MEMBARRIER
197 enum {
198 	MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY		= (1U << 0),
199 	MEMBARRIER_STATE_PRIVATE_EXPEDITED			= (1U << 1),
200 	MEMBARRIER_STATE_GLOBAL_EXPEDITED_READY			= (1U << 2),
201 	MEMBARRIER_STATE_GLOBAL_EXPEDITED			= (1U << 3),
202 	MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE_READY	= (1U << 4),
203 	MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE		= (1U << 5),
204 };
205 
206 enum {
207 	MEMBARRIER_FLAG_SYNC_CORE	= (1U << 0),
208 };
209 
210 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
211 #include <asm/membarrier.h>
212 #endif
213 
214 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
215 {
216 	if (likely(!(atomic_read(&mm->membarrier_state) &
217 		     MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE)))
218 		return;
219 	sync_core_before_usermode();
220 }
221 
222 static inline void membarrier_execve(struct task_struct *t)
223 {
224 	atomic_set(&t->mm->membarrier_state, 0);
225 }
226 #else
227 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
228 static inline void membarrier_arch_switch_mm(struct mm_struct *prev,
229 					     struct mm_struct *next,
230 					     struct task_struct *tsk)
231 {
232 }
233 #endif
234 static inline void membarrier_execve(struct task_struct *t)
235 {
236 }
237 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
238 {
239 }
240 #endif
241 
242 #endif /* _LINUX_SCHED_MM_H */
243