xref: /linux-6.15/include/linux/sched/mm.h (revision ce63b2c8)
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 static inline void mmdrop(struct mm_struct *mm)
42 {
43 	/*
44 	 * The implicit full barrier implied by atomic_dec_and_test() is
45 	 * required by the membarrier system call before returning to
46 	 * user-space, after storing to rq->curr.
47 	 */
48 	if (unlikely(atomic_dec_and_test(&mm->mm_count)))
49 		__mmdrop(mm);
50 }
51 
52 /**
53  * mmget() - Pin the address space associated with a &struct mm_struct.
54  * @mm: The address space to pin.
55  *
56  * Make sure that the address space of the given &struct mm_struct doesn't
57  * go away. This does not protect against parts of the address space being
58  * modified or freed, however.
59  *
60  * Never use this function to pin this address space for an
61  * unbounded/indefinite amount of time.
62  *
63  * Use mmput() to release the reference acquired by mmget().
64  *
65  * See also <Documentation/vm/active_mm.txt> for an in-depth explanation
66  * of &mm_struct.mm_count vs &mm_struct.mm_users.
67  */
68 static inline void mmget(struct mm_struct *mm)
69 {
70 	atomic_inc(&mm->mm_users);
71 }
72 
73 static inline bool mmget_not_zero(struct mm_struct *mm)
74 {
75 	return atomic_inc_not_zero(&mm->mm_users);
76 }
77 
78 /* mmput gets rid of the mappings and all user-space */
79 extern void mmput(struct mm_struct *);
80 #ifdef CONFIG_MMU
81 /* same as above but performs the slow path from the async context. Can
82  * be called from the atomic context as well
83  */
84 void mmput_async(struct mm_struct *);
85 #endif
86 
87 /* Grab a reference to a task's mm, if it is not already going away */
88 extern struct mm_struct *get_task_mm(struct task_struct *task);
89 /*
90  * Grab a reference to a task's mm, if it is not already going away
91  * and ptrace_may_access with the mode parameter passed to it
92  * succeeds.
93  */
94 extern struct mm_struct *mm_access(struct task_struct *task, unsigned int mode);
95 /* Remove the current tasks stale references to the old mm_struct */
96 extern void mm_release(struct task_struct *, struct mm_struct *);
97 
98 #ifdef CONFIG_MEMCG
99 extern void mm_update_next_owner(struct mm_struct *mm);
100 #else
101 static inline void mm_update_next_owner(struct mm_struct *mm)
102 {
103 }
104 #endif /* CONFIG_MEMCG */
105 
106 #ifdef CONFIG_MMU
107 extern void arch_pick_mmap_layout(struct mm_struct *mm,
108 				  struct rlimit *rlim_stack);
109 extern unsigned long
110 arch_get_unmapped_area(struct file *, unsigned long, unsigned long,
111 		       unsigned long, unsigned long);
112 extern unsigned long
113 arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
114 			  unsigned long len, unsigned long pgoff,
115 			  unsigned long flags);
116 #else
117 static inline void arch_pick_mmap_layout(struct mm_struct *mm,
118 					 struct rlimit *rlim_stack) {}
119 #endif
120 
121 static inline bool in_vfork(struct task_struct *tsk)
122 {
123 	bool ret;
124 
125 	/*
126 	 * need RCU to access ->real_parent if CLONE_VM was used along with
127 	 * CLONE_PARENT.
128 	 *
129 	 * We check real_parent->mm == tsk->mm because CLONE_VFORK does not
130 	 * imply CLONE_VM
131 	 *
132 	 * CLONE_VFORK can be used with CLONE_PARENT/CLONE_THREAD and thus
133 	 * ->real_parent is not necessarily the task doing vfork(), so in
134 	 * theory we can't rely on task_lock() if we want to dereference it.
135 	 *
136 	 * And in this case we can't trust the real_parent->mm == tsk->mm
137 	 * check, it can be false negative. But we do not care, if init or
138 	 * another oom-unkillable task does this it should blame itself.
139 	 */
140 	rcu_read_lock();
141 	ret = tsk->vfork_done && tsk->real_parent->mm == tsk->mm;
142 	rcu_read_unlock();
143 
144 	return ret;
145 }
146 
147 /*
148  * Applies per-task gfp context to the given allocation flags.
149  * PF_MEMALLOC_NOIO implies GFP_NOIO
150  * PF_MEMALLOC_NOFS implies GFP_NOFS
151  */
152 static inline gfp_t current_gfp_context(gfp_t flags)
153 {
154 	/*
155 	 * NOIO implies both NOIO and NOFS and it is a weaker context
156 	 * so always make sure it makes precendence
157 	 */
158 	if (unlikely(current->flags & PF_MEMALLOC_NOIO))
159 		flags &= ~(__GFP_IO | __GFP_FS);
160 	else if (unlikely(current->flags & PF_MEMALLOC_NOFS))
161 		flags &= ~__GFP_FS;
162 	return flags;
163 }
164 
165 #ifdef CONFIG_LOCKDEP
166 extern void fs_reclaim_acquire(gfp_t gfp_mask);
167 extern void fs_reclaim_release(gfp_t gfp_mask);
168 #else
169 static inline void fs_reclaim_acquire(gfp_t gfp_mask) { }
170 static inline void fs_reclaim_release(gfp_t gfp_mask) { }
171 #endif
172 
173 static inline unsigned int memalloc_noio_save(void)
174 {
175 	unsigned int flags = current->flags & PF_MEMALLOC_NOIO;
176 	current->flags |= PF_MEMALLOC_NOIO;
177 	return flags;
178 }
179 
180 static inline void memalloc_noio_restore(unsigned int flags)
181 {
182 	current->flags = (current->flags & ~PF_MEMALLOC_NOIO) | flags;
183 }
184 
185 static inline unsigned int memalloc_nofs_save(void)
186 {
187 	unsigned int flags = current->flags & PF_MEMALLOC_NOFS;
188 	current->flags |= PF_MEMALLOC_NOFS;
189 	return flags;
190 }
191 
192 static inline void memalloc_nofs_restore(unsigned int flags)
193 {
194 	current->flags = (current->flags & ~PF_MEMALLOC_NOFS) | flags;
195 }
196 
197 static inline unsigned int memalloc_noreclaim_save(void)
198 {
199 	unsigned int flags = current->flags & PF_MEMALLOC;
200 	current->flags |= PF_MEMALLOC;
201 	return flags;
202 }
203 
204 static inline void memalloc_noreclaim_restore(unsigned int flags)
205 {
206 	current->flags = (current->flags & ~PF_MEMALLOC) | flags;
207 }
208 
209 #ifdef CONFIG_MEMBARRIER
210 enum {
211 	MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY		= (1U << 0),
212 	MEMBARRIER_STATE_PRIVATE_EXPEDITED			= (1U << 1),
213 	MEMBARRIER_STATE_GLOBAL_EXPEDITED_READY			= (1U << 2),
214 	MEMBARRIER_STATE_GLOBAL_EXPEDITED			= (1U << 3),
215 	MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE_READY	= (1U << 4),
216 	MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE		= (1U << 5),
217 };
218 
219 enum {
220 	MEMBARRIER_FLAG_SYNC_CORE	= (1U << 0),
221 };
222 
223 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
224 #include <asm/membarrier.h>
225 #endif
226 
227 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
228 {
229 	if (likely(!(atomic_read(&mm->membarrier_state) &
230 		     MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE)))
231 		return;
232 	sync_core_before_usermode();
233 }
234 
235 static inline void membarrier_execve(struct task_struct *t)
236 {
237 	atomic_set(&t->mm->membarrier_state, 0);
238 }
239 #else
240 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
241 static inline void membarrier_arch_switch_mm(struct mm_struct *prev,
242 					     struct mm_struct *next,
243 					     struct task_struct *tsk)
244 {
245 }
246 #endif
247 static inline void membarrier_execve(struct task_struct *t)
248 {
249 }
250 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
251 {
252 }
253 #endif
254 
255 #endif /* _LINUX_SCHED_MM_H */
256