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 * PF_MEMALLOC_NOCMA implies no allocation from CMA region.
152 */
153static inline gfp_t current_gfp_context(gfp_t flags)
154{
155 if (unlikely(current->flags &
156 (PF_MEMALLOC_NOIO | PF_MEMALLOC_NOFS | PF_MEMALLOC_NOCMA))) {
157 /*
158 * NOIO implies both NOIO and NOFS and it is a weaker context
159 * so always make sure it makes precedence
160 */
161 if (current->flags & PF_MEMALLOC_NOIO)
162 flags &= ~(__GFP_IO | __GFP_FS);
163 else if (current->flags & PF_MEMALLOC_NOFS)
164 flags &= ~__GFP_FS;
165#ifdef CONFIG_CMA
166 if (current->flags & PF_MEMALLOC_NOCMA)
167 flags &= ~__GFP_MOVABLE;
168#endif
169 }
170 return flags;
171}
172
173#ifdef CONFIG_LOCKDEP
174extern void __fs_reclaim_acquire(void);
175extern void __fs_reclaim_release(void);
176extern void fs_reclaim_acquire(gfp_t gfp_mask);
177extern void fs_reclaim_release(gfp_t gfp_mask);
178#else
179static inline void __fs_reclaim_acquire(void) { }
180static inline void __fs_reclaim_release(void) { }
181static inline void fs_reclaim_acquire(gfp_t gfp_mask) { }
182static inline void fs_reclaim_release(gfp_t gfp_mask) { }
183#endif
184
185/**
186 * memalloc_noio_save - Marks implicit GFP_NOIO allocation scope.
187 *
188 * This functions marks the beginning of the GFP_NOIO allocation scope.
189 * All further allocations will implicitly drop __GFP_IO flag and so
190 * they are safe for the IO critical section from the allocation recursion
191 * point of view. Use memalloc_noio_restore to end the scope with flags
192 * returned by this function.
193 *
194 * This function is safe to be used from any context.
195 */
196static inline unsigned int memalloc_noio_save(void)
197{
198 unsigned int flags = current->flags & PF_MEMALLOC_NOIO;
199 current->flags |= PF_MEMALLOC_NOIO;
200 return flags;
201}
202
203/**
204 * memalloc_noio_restore - Ends the implicit GFP_NOIO scope.
205 * @flags: Flags to restore.
206 *
207 * Ends the implicit GFP_NOIO scope started by memalloc_noio_save function.
208 * Always make sure that that the given flags is the return value from the
209 * pairing memalloc_noio_save call.
210 */
211static inline void memalloc_noio_restore(unsigned int flags)
212{
213 current->flags = (current->flags & ~PF_MEMALLOC_NOIO) | flags;
214}
215
216/**
217 * memalloc_nofs_save - Marks implicit GFP_NOFS allocation scope.
218 *
219 * This functions marks the beginning of the GFP_NOFS allocation scope.
220 * All further allocations will implicitly drop __GFP_FS flag and so
221 * they are safe for the FS critical section from the allocation recursion
222 * point of view. Use memalloc_nofs_restore to end the scope with flags
223 * returned by this function.
224 *
225 * This function is safe to be used from any context.
226 */
227static inline unsigned int memalloc_nofs_save(void)
228{
229 unsigned int flags = current->flags & PF_MEMALLOC_NOFS;
230 current->flags |= PF_MEMALLOC_NOFS;
231 return flags;
232}
233
234/**
235 * memalloc_nofs_restore - Ends the implicit GFP_NOFS scope.
236 * @flags: Flags to restore.
237 *
238 * Ends the implicit GFP_NOFS scope started by memalloc_nofs_save function.
239 * Always make sure that that the given flags is the return value from the
240 * pairing memalloc_nofs_save call.
241 */
242static inline void memalloc_nofs_restore(unsigned int flags)
243{
244 current->flags = (current->flags & ~PF_MEMALLOC_NOFS) | flags;
245}
246
247static inline unsigned int memalloc_noreclaim_save(void)
248{
249 unsigned int flags = current->flags & PF_MEMALLOC;
250 current->flags |= PF_MEMALLOC;
251 return flags;
252}
253
254static inline void memalloc_noreclaim_restore(unsigned int flags)
255{
256 current->flags = (current->flags & ~PF_MEMALLOC) | flags;
257}
258
259#ifdef CONFIG_CMA
260static inline unsigned int memalloc_nocma_save(void)
261{
262 unsigned int flags = current->flags & PF_MEMALLOC_NOCMA;
263
264 current->flags |= PF_MEMALLOC_NOCMA;
265 return flags;
266}
267
268static inline void memalloc_nocma_restore(unsigned int flags)
269{
270 current->flags = (current->flags & ~PF_MEMALLOC_NOCMA) | flags;
271}
272#else
273static inline unsigned int memalloc_nocma_save(void)
274{
275 return 0;
276}
277
278static inline void memalloc_nocma_restore(unsigned int flags)
279{
280}
281#endif
282
283#ifdef CONFIG_MEMCG
284/**
285 * memalloc_use_memcg - Starts the remote memcg charging scope.
286 * @memcg: memcg to charge.
287 *
288 * This function marks the beginning of the remote memcg charging scope. All the
289 * __GFP_ACCOUNT allocations till the end of the scope will be charged to the
290 * given memcg.
291 *
292 * NOTE: This function is not nesting safe.
293 */
294static inline void memalloc_use_memcg(struct mem_cgroup *memcg)
295{
296 WARN_ON_ONCE(current->active_memcg);
297 current->active_memcg = memcg;
298}
299
300/**
301 * memalloc_unuse_memcg - Ends the remote memcg charging scope.
302 *
303 * This function marks the end of the remote memcg charging scope started by
304 * memalloc_use_memcg().
305 */
306static inline void memalloc_unuse_memcg(void)
307{
308 current->active_memcg = NULL;
309}
310#else
311static inline void memalloc_use_memcg(struct mem_cgroup *memcg)
312{
313}
314
315static inline void memalloc_unuse_memcg(void)
316{
317}
318#endif
319
320#ifdef CONFIG_MEMBARRIER
321enum {
322 MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY = (1U << 0),
323 MEMBARRIER_STATE_PRIVATE_EXPEDITED = (1U << 1),
324 MEMBARRIER_STATE_GLOBAL_EXPEDITED_READY = (1U << 2),
325 MEMBARRIER_STATE_GLOBAL_EXPEDITED = (1U << 3),
326 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE_READY = (1U << 4),
327 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE = (1U << 5),
328};
329
330enum {
331 MEMBARRIER_FLAG_SYNC_CORE = (1U << 0),
332};
333
334#ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
335#include <asm/membarrier.h>
336#endif
337
338static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
339{
340 if (likely(!(atomic_read(&mm->membarrier_state) &
341 MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE)))
342 return;
343 sync_core_before_usermode();
344}
345
346static 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
352static inline void membarrier_arch_switch_mm(struct mm_struct *prev,
353 struct mm_struct *next,
354 struct task_struct *tsk)
355{
356}
357#endif
358static inline void membarrier_execve(struct task_struct *t)
359{
360}
361static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
362{
363}
364#endif
365
366#endif /* _LINUX_SCHED_MM_H */
367