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.rst> for an in-depth explanation
32  * of &mm_struct.mm_count vs &mm_struct.mm_users.
33  */
mmgrab(struct mm_struct * mm)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 
mmdrop(struct mm_struct * mm)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 void mmdrop(struct mm_struct *mm);
53 
54 /*
55  * This has to be called after a get_task_mm()/mmget_not_zero()
56  * followed by taking the mmap_sem for writing before modifying the
57  * vmas or anything the coredump pretends not to change from under it.
58  *
59  * It also has to be called when mmgrab() is used in the context of
60  * the process, but then the mm_count refcount is transferred outside
61  * the context of the process to run down_write() on that pinned mm.
62  *
63  * NOTE: find_extend_vma() called from GUP context is the only place
64  * that can modify the "mm" (notably the vm_start/end) under mmap_sem
65  * for reading and outside the context of the process, so it is also
66  * the only case that holds the mmap_sem for reading that must call
67  * this function. Generally if the mmap_sem is hold for reading
68  * there's no need of this check after get_task_mm()/mmget_not_zero().
69  *
70  * This function can be obsoleted and the check can be removed, after
71  * the coredump code will hold the mmap_sem for writing before
72  * invoking the ->core_dump methods.
73  */
mmget_still_valid(struct mm_struct * mm)74 static inline bool mmget_still_valid(struct mm_struct *mm)
75 {
76 	return likely(!mm->core_state);
77 }
78 
79 /**
80  * mmget() - Pin the address space associated with a &struct mm_struct.
81  * @mm: The address space to pin.
82  *
83  * Make sure that the address space of the given &struct mm_struct doesn't
84  * go away. This does not protect against parts of the address space being
85  * modified or freed, however.
86  *
87  * Never use this function to pin this address space for an
88  * unbounded/indefinite amount of time.
89  *
90  * Use mmput() to release the reference acquired by mmget().
91  *
92  * See also <Documentation/vm/active_mm.rst> for an in-depth explanation
93  * of &mm_struct.mm_count vs &mm_struct.mm_users.
94  */
mmget(struct mm_struct * mm)95 static inline void mmget(struct mm_struct *mm)
96 {
97 	atomic_inc(&mm->mm_users);
98 }
99 
mmget_not_zero(struct mm_struct * mm)100 static inline bool mmget_not_zero(struct mm_struct *mm)
101 {
102 	return atomic_inc_not_zero(&mm->mm_users);
103 }
104 
105 /* mmput gets rid of the mappings and all user-space */
106 extern void mmput(struct mm_struct *);
107 #ifdef CONFIG_MMU
108 /* same as above but performs the slow path from the async context. Can
109  * be called from the atomic context as well
110  */
111 void mmput_async(struct mm_struct *);
112 #endif
113 
114 /* Grab a reference to a task's mm, if it is not already going away */
115 extern struct mm_struct *get_task_mm(struct task_struct *task);
116 /*
117  * Grab a reference to a task's mm, if it is not already going away
118  * and ptrace_may_access with the mode parameter passed to it
119  * succeeds.
120  */
121 extern struct mm_struct *mm_access(struct task_struct *task, unsigned int mode);
122 /* Remove the current tasks stale references to the old mm_struct on exit() */
123 extern void exit_mm_release(struct task_struct *, struct mm_struct *);
124 /* Remove the current tasks stale references to the old mm_struct on exec() */
125 extern void exec_mm_release(struct task_struct *, struct mm_struct *);
126 
127 #ifdef CONFIG_MEMCG
128 extern void mm_update_next_owner(struct mm_struct *mm);
129 #else
mm_update_next_owner(struct mm_struct * mm)130 static inline void mm_update_next_owner(struct mm_struct *mm)
131 {
132 }
133 #endif /* CONFIG_MEMCG */
134 
135 #ifdef CONFIG_MMU
136 extern void arch_pick_mmap_layout(struct mm_struct *mm,
137 				  struct rlimit *rlim_stack);
138 extern unsigned long
139 arch_get_unmapped_area(struct file *, unsigned long, unsigned long,
140 		       unsigned long, unsigned long);
141 extern unsigned long
142 arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
143 			  unsigned long len, unsigned long pgoff,
144 			  unsigned long flags);
145 #else
arch_pick_mmap_layout(struct mm_struct * mm,struct rlimit * rlim_stack)146 static inline void arch_pick_mmap_layout(struct mm_struct *mm,
147 					 struct rlimit *rlim_stack) {}
148 #endif
149 
in_vfork(struct task_struct * tsk)150 static inline bool in_vfork(struct task_struct *tsk)
151 {
152 	bool ret;
153 
154 	/*
155 	 * need RCU to access ->real_parent if CLONE_VM was used along with
156 	 * CLONE_PARENT.
157 	 *
158 	 * We check real_parent->mm == tsk->mm because CLONE_VFORK does not
159 	 * imply CLONE_VM
160 	 *
161 	 * CLONE_VFORK can be used with CLONE_PARENT/CLONE_THREAD and thus
162 	 * ->real_parent is not necessarily the task doing vfork(), so in
163 	 * theory we can't rely on task_lock() if we want to dereference it.
164 	 *
165 	 * And in this case we can't trust the real_parent->mm == tsk->mm
166 	 * check, it can be false negative. But we do not care, if init or
167 	 * another oom-unkillable task does this it should blame itself.
168 	 */
169 	rcu_read_lock();
170 	ret = tsk->vfork_done &&
171 			rcu_dereference(tsk->real_parent)->mm == tsk->mm;
172 	rcu_read_unlock();
173 
174 	return ret;
175 }
176 
177 /*
178  * Applies per-task gfp context to the given allocation flags.
179  * PF_MEMALLOC_NOIO implies GFP_NOIO
180  * PF_MEMALLOC_NOFS implies GFP_NOFS
181  */
current_gfp_context(gfp_t flags)182 static inline gfp_t current_gfp_context(gfp_t flags)
183 {
184 	/*
185 	 * NOIO implies both NOIO and NOFS and it is a weaker context
186 	 * so always make sure it makes precendence
187 	 */
188 	if (unlikely(current->flags & PF_MEMALLOC_NOIO))
189 		flags &= ~(__GFP_IO | __GFP_FS);
190 	else if (unlikely(current->flags & PF_MEMALLOC_NOFS))
191 		flags &= ~__GFP_FS;
192 	return flags;
193 }
194 
195 #ifdef CONFIG_LOCKDEP
196 extern void __fs_reclaim_acquire(void);
197 extern void __fs_reclaim_release(void);
198 extern void fs_reclaim_acquire(gfp_t gfp_mask);
199 extern void fs_reclaim_release(gfp_t gfp_mask);
200 #else
__fs_reclaim_acquire(void)201 static inline void __fs_reclaim_acquire(void) { }
__fs_reclaim_release(void)202 static inline void __fs_reclaim_release(void) { }
fs_reclaim_acquire(gfp_t gfp_mask)203 static inline void fs_reclaim_acquire(gfp_t gfp_mask) { }
fs_reclaim_release(gfp_t gfp_mask)204 static inline void fs_reclaim_release(gfp_t gfp_mask) { }
205 #endif
206 
207 /**
208  * memalloc_noio_save - Marks implicit GFP_NOIO allocation scope.
209  *
210  * This functions marks the beginning of the GFP_NOIO allocation scope.
211  * All further allocations will implicitly drop __GFP_IO flag and so
212  * they are safe for the IO critical section from the allocation recursion
213  * point of view. Use memalloc_noio_restore to end the scope with flags
214  * returned by this function.
215  *
216  * This function is safe to be used from any context.
217  */
memalloc_noio_save(void)218 static inline unsigned int memalloc_noio_save(void)
219 {
220 	unsigned int flags = current->flags & PF_MEMALLOC_NOIO;
221 	current->flags |= PF_MEMALLOC_NOIO;
222 	return flags;
223 }
224 
225 /**
226  * memalloc_noio_restore - Ends the implicit GFP_NOIO scope.
227  * @flags: Flags to restore.
228  *
229  * Ends the implicit GFP_NOIO scope started by memalloc_noio_save function.
230  * Always make sure that that the given flags is the return value from the
231  * pairing memalloc_noio_save call.
232  */
memalloc_noio_restore(unsigned int flags)233 static inline void memalloc_noio_restore(unsigned int flags)
234 {
235 	current->flags = (current->flags & ~PF_MEMALLOC_NOIO) | flags;
236 }
237 
238 /**
239  * memalloc_nofs_save - Marks implicit GFP_NOFS allocation scope.
240  *
241  * This functions marks the beginning of the GFP_NOFS allocation scope.
242  * All further allocations will implicitly drop __GFP_FS flag and so
243  * they are safe for the FS critical section from the allocation recursion
244  * point of view. Use memalloc_nofs_restore to end the scope with flags
245  * returned by this function.
246  *
247  * This function is safe to be used from any context.
248  */
memalloc_nofs_save(void)249 static inline unsigned int memalloc_nofs_save(void)
250 {
251 	unsigned int flags = current->flags & PF_MEMALLOC_NOFS;
252 	current->flags |= PF_MEMALLOC_NOFS;
253 	return flags;
254 }
255 
256 /**
257  * memalloc_nofs_restore - Ends the implicit GFP_NOFS scope.
258  * @flags: Flags to restore.
259  *
260  * Ends the implicit GFP_NOFS scope started by memalloc_nofs_save function.
261  * Always make sure that that the given flags is the return value from the
262  * pairing memalloc_nofs_save call.
263  */
memalloc_nofs_restore(unsigned int flags)264 static inline void memalloc_nofs_restore(unsigned int flags)
265 {
266 	current->flags = (current->flags & ~PF_MEMALLOC_NOFS) | flags;
267 }
268 
memalloc_noreclaim_save(void)269 static inline unsigned int memalloc_noreclaim_save(void)
270 {
271 	unsigned int flags = current->flags & PF_MEMALLOC;
272 	current->flags |= PF_MEMALLOC;
273 	return flags;
274 }
275 
memalloc_noreclaim_restore(unsigned int flags)276 static inline void memalloc_noreclaim_restore(unsigned int flags)
277 {
278 	current->flags = (current->flags & ~PF_MEMALLOC) | flags;
279 }
280 
281 #ifdef CONFIG_MEMCG
282 /**
283  * memalloc_use_memcg - Starts the remote memcg charging scope.
284  * @memcg: memcg to charge.
285  *
286  * This function marks the beginning of the remote memcg charging scope. All the
287  * __GFP_ACCOUNT allocations till the end of the scope will be charged to the
288  * given memcg.
289  *
290  * NOTE: This function is not nesting safe.
291  */
memalloc_use_memcg(struct mem_cgroup * memcg)292 static inline void memalloc_use_memcg(struct mem_cgroup *memcg)
293 {
294 	WARN_ON_ONCE(current->active_memcg);
295 	current->active_memcg = memcg;
296 }
297 
298 /**
299  * memalloc_unuse_memcg - Ends the remote memcg charging scope.
300  *
301  * This function marks the end of the remote memcg charging scope started by
302  * memalloc_use_memcg().
303  */
memalloc_unuse_memcg(void)304 static inline void memalloc_unuse_memcg(void)
305 {
306 	current->active_memcg = NULL;
307 }
308 #else
memalloc_use_memcg(struct mem_cgroup * memcg)309 static inline void memalloc_use_memcg(struct mem_cgroup *memcg)
310 {
311 }
312 
memalloc_unuse_memcg(void)313 static inline void memalloc_unuse_memcg(void)
314 {
315 }
316 #endif
317 
318 #ifdef CONFIG_MEMBARRIER
319 enum {
320 	MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY		= (1U << 0),
321 	MEMBARRIER_STATE_PRIVATE_EXPEDITED			= (1U << 1),
322 	MEMBARRIER_STATE_GLOBAL_EXPEDITED_READY			= (1U << 2),
323 	MEMBARRIER_STATE_GLOBAL_EXPEDITED			= (1U << 3),
324 	MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE_READY	= (1U << 4),
325 	MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE		= (1U << 5),
326 };
327 
328 enum {
329 	MEMBARRIER_FLAG_SYNC_CORE	= (1U << 0),
330 };
331 
332 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
333 #include <asm/membarrier.h>
334 #endif
335 
membarrier_mm_sync_core_before_usermode(struct mm_struct * mm)336 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
337 {
338 	if (current->mm != mm)
339 		return;
340 	if (likely(!(atomic_read(&mm->membarrier_state) &
341 		     MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE)))
342 		return;
343 	sync_core_before_usermode();
344 }
345 
membarrier_execve(struct task_struct * t)346 static inline void membarrier_execve(struct task_struct *t)
347 {
348 	atomic_set(&t->mm->membarrier_state, 0);
349 }
350 #else
351 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
membarrier_arch_switch_mm(struct mm_struct * prev,struct mm_struct * next,struct task_struct * tsk)352 static inline void membarrier_arch_switch_mm(struct mm_struct *prev,
353 					     struct mm_struct *next,
354 					     struct task_struct *tsk)
355 {
356 }
357 #endif
membarrier_execve(struct task_struct * t)358 static inline void membarrier_execve(struct task_struct *t)
359 {
360 }
membarrier_mm_sync_core_before_usermode(struct mm_struct * mm)361 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
362 {
363 }
364 #endif
365 
366 #endif /* _LINUX_SCHED_MM_H */
367