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 */
15extern 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 */
34static inline void mmgrab(struct mm_struct *mm)
35{
36 atomic_inc(&mm->mm_count);
37}
38
39extern void __mmdrop(struct mm_struct *mm);
40
41static 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.rst> for an in-depth explanation
66 * of &mm_struct.mm_count vs &mm_struct.mm_users.
67 */
68static inline void mmget(struct mm_struct *mm)
69{
70 atomic_inc(&mm->mm_users);
71}
72
73static 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 */
79extern 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 */
84void mmput_async(struct mm_struct *);
85#endif
86
87/* Grab a reference to a task's mm, if it is not already going away */
88extern 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 */
94extern struct mm_struct *mm_access(struct task_struct *task, unsigned int mode);
95/* Remove the current tasks stale references to the old mm_struct */
96extern void mm_release(struct task_struct *, struct mm_struct *);
97
98#ifdef CONFIG_MEMCG
99extern void mm_update_next_owner(struct mm_struct *mm);
100#else
101static inline void mm_update_next_owner(struct mm_struct *mm)
102{
103}
104#endif /* CONFIG_MEMCG */
105
106#ifdef CONFIG_MMU
107extern void arch_pick_mmap_layout(struct mm_struct *mm,
108 struct rlimit *rlim_stack);
109extern unsigned long
110arch_get_unmapped_area(struct file *, unsigned long, unsigned long,
111 unsigned long, unsigned long);
112extern unsigned long
113arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
114 unsigned long len, unsigned long pgoff,
115 unsigned long flags);
116#else
117static inline void arch_pick_mmap_layout(struct mm_struct *mm,
118 struct rlimit *rlim_stack) {}
119#endif
120
121static 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 */
152static 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
166extern void __fs_reclaim_acquire(void);
167extern void __fs_reclaim_release(void);
168extern void fs_reclaim_acquire(gfp_t gfp_mask);
169extern void fs_reclaim_release(gfp_t gfp_mask);
170#else
171static inline void __fs_reclaim_acquire(void) { }
172static inline void __fs_reclaim_release(void) { }
173static inline void fs_reclaim_acquire(gfp_t gfp_mask) { }
174static inline void fs_reclaim_release(gfp_t gfp_mask) { }
175#endif
176
177/**
178 * memalloc_noio_save - Marks implicit GFP_NOIO allocation scope.
179 *
180 * This functions marks the beginning of the GFP_NOIO allocation scope.
181 * All further allocations will implicitly drop __GFP_IO flag and so
182 * they are safe for the IO critical section from the allocation recursion
183 * point of view. Use memalloc_noio_restore to end the scope with flags
184 * returned by this function.
185 *
186 * This function is safe to be used from any context.
187 */
188static inline unsigned int memalloc_noio_save(void)
189{
190 unsigned int flags = current->flags & PF_MEMALLOC_NOIO;
191 current->flags |= PF_MEMALLOC_NOIO;
192 return flags;
193}
194
195/**
196 * memalloc_noio_restore - Ends the implicit GFP_NOIO scope.
197 * @flags: Flags to restore.
198 *
199 * Ends the implicit GFP_NOIO scope started by memalloc_noio_save function.
200 * Always make sure that that the given flags is the return value from the
201 * pairing memalloc_noio_save call.
202 */
203static inline void memalloc_noio_restore(unsigned int flags)
204{
205 current->flags = (current->flags & ~PF_MEMALLOC_NOIO) | flags;
206}
207
208/**
209 * memalloc_nofs_save - Marks implicit GFP_NOFS allocation scope.
210 *
211 * This functions marks the beginning of the GFP_NOFS allocation scope.
212 * All further allocations will implicitly drop __GFP_FS flag and so
213 * they are safe for the FS critical section from the allocation recursion
214 * point of view. Use memalloc_nofs_restore to end the scope with flags
215 * returned by this function.
216 *
217 * This function is safe to be used from any context.
218 */
219static inline unsigned int memalloc_nofs_save(void)
220{
221 unsigned int flags = current->flags & PF_MEMALLOC_NOFS;
222 current->flags |= PF_MEMALLOC_NOFS;
223 return flags;
224}
225
226/**
227 * memalloc_nofs_restore - Ends the implicit GFP_NOFS scope.
228 * @flags: Flags to restore.
229 *
230 * Ends the implicit GFP_NOFS scope started by memalloc_nofs_save function.
231 * Always make sure that that the given flags is the return value from the
232 * pairing memalloc_nofs_save call.
233 */
234static inline void memalloc_nofs_restore(unsigned int flags)
235{
236 current->flags = (current->flags & ~PF_MEMALLOC_NOFS) | flags;
237}
238
239static inline unsigned int memalloc_noreclaim_save(void)
240{
241 unsigned int flags = current->flags & PF_MEMALLOC;
242 current->flags |= PF_MEMALLOC;
243 return flags;
244}
245
246static inline void memalloc_noreclaim_restore(unsigned int flags)
247{
248 current->flags = (current->flags & ~PF_MEMALLOC) | flags;
249}
250
251#ifdef CONFIG_MEMCG
252/**
253 * memalloc_use_memcg - Starts the remote memcg charging scope.
254 * @memcg: memcg to charge.
255 *
256 * This function marks the beginning of the remote memcg charging scope. All the
257 * __GFP_ACCOUNT allocations till the end of the scope will be charged to the
258 * given memcg.
259 *
260 * NOTE: This function is not nesting safe.
261 */
262static inline void memalloc_use_memcg(struct mem_cgroup *memcg)
263{
264 WARN_ON_ONCE(current->active_memcg);
265 current->active_memcg = memcg;
266}
267
268/**
269 * memalloc_unuse_memcg - Ends the remote memcg charging scope.
270 *
271 * This function marks the end of the remote memcg charging scope started by
272 * memalloc_use_memcg().
273 */
274static inline void memalloc_unuse_memcg(void)
275{
276 current->active_memcg = NULL;
277}
278#else
279static inline void memalloc_use_memcg(struct mem_cgroup *memcg)
280{
281}
282
283static inline void memalloc_unuse_memcg(void)
284{
285}
286#endif
287
288#ifdef CONFIG_MEMBARRIER
289enum {
290 MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY = (1U << 0),
291 MEMBARRIER_STATE_PRIVATE_EXPEDITED = (1U << 1),
292 MEMBARRIER_STATE_GLOBAL_EXPEDITED_READY = (1U << 2),
293 MEMBARRIER_STATE_GLOBAL_EXPEDITED = (1U << 3),
294 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE_READY = (1U << 4),
295 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE = (1U << 5),
296};
297
298enum {
299 MEMBARRIER_FLAG_SYNC_CORE = (1U << 0),
300};
301
302#ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
303#include <asm/membarrier.h>
304#endif
305
306static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
307{
308 if (likely(!(atomic_read(&mm->membarrier_state) &
309 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE)))
310 return;
311 sync_core_before_usermode();
312}
313
314static inline void membarrier_execve(struct task_struct *t)
315{
316 atomic_set(&t->mm->membarrier_state, 0);
317}
318#else
319#ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
320static inline void membarrier_arch_switch_mm(struct mm_struct *prev,
321 struct mm_struct *next,
322 struct task_struct *tsk)
323{
324}
325#endif
326static inline void membarrier_execve(struct task_struct *t)
327{
328}
329static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
330{
331}
332#endif
333
334#endif /* _LINUX_SCHED_MM_H */
335