include/linux/sched/mm.h

   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
  11 /*
  12  * Routines for handling mm_structs
  13  */
  14 extern struct mm_struct * mm_alloc(void);
  15
  16 /**
  17  * mmgrab() - Pin a &struct mm_struct.
  18  * @mm: The &struct mm_struct to pin.
  19  *
  20  * Make sure that @mm will not get freed even after the owning task
  21  * exits. This doesn't guarantee that the associated address space
  22  * will still exist later on and mmget_not_zero() has to be used before
  23  * accessing it.
  24  *
  25  * This is a preferred way to to pin @mm for a longer/unbounded amount
  26  * of time.
  27  *
  28  * Use mmdrop() to release the reference acquired by mmgrab().
  29  *
  30  * See also <Documentation/vm/active_mm.txt> for an in-depth explanation
  31  * of &mm_struct.mm_count vs &mm_struct.mm_users.
  32  */
  33 static inline void mmgrab(struct mm_struct *mm)
  34 {
  35         atomic_inc(&mm->mm_count);
  36 }
  37
  38 /* mmdrop drops the mm and the page tables */
  39 extern void __mmdrop(struct mm_struct *);
  40 static inline void mmdrop(struct mm_struct *mm)
  41 {
  42         if (unlikely(atomic_dec_and_test(&mm->mm_count)))
  43                 __mmdrop(mm);
  44 }
  45
  46 static inline void mmdrop_async_fn(struct work_struct *work)
  47 {
  48         struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
  49         __mmdrop(mm);
  50 }
  51
  52 static inline void mmdrop_async(struct mm_struct *mm)
  53 {
  54         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
  55                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
  56                 schedule_work(&mm->async_put_work);
  57         }
  58 }
  59
  60 /*
  61  * This has to be called after a get_task_mm()/mmget_not_zero()
  62  * followed by taking the mmap_sem for writing before modifying the
  63  * vmas or anything the coredump pretends not to change from under it.
  64  *
  65  * NOTE: find_extend_vma() called from GUP context is the only place
  66  * that can modify the "mm" (notably the vm_start/end) under mmap_sem
  67  * for reading and outside the context of the process, so it is also
  68  * the only case that holds the mmap_sem for reading that must call
  69  * this function. Generally if the mmap_sem is hold for reading
  70  * there's no need of this check after get_task_mm()/mmget_not_zero().
  71  *
  72  * This function can be obsoleted and the check can be removed, after
  73  * the coredump code will hold the mmap_sem for writing before
  74  * invoking the ->core_dump methods.
  75  */
  76 static inline bool mmget_still_valid(struct mm_struct *mm)
  77 {
  78         return likely(!mm->core_state);
  79 }
  80
  81 /**
  82  * mmget() - Pin the address space associated with a &struct mm_struct.
  83  * @mm: The address space to pin.
  84  *
  85  * Make sure that the address space of the given &struct mm_struct doesn't
  86  * go away. This does not protect against parts of the address space being
  87  * modified or freed, however.
  88  *
  89  * Never use this function to pin this address space for an
  90  * unbounded/indefinite amount of time.
  91  *
  92  * Use mmput() to release the reference acquired by mmget().
  93  *
  94  * See also <Documentation/vm/active_mm.txt> for an in-depth explanation
  95  * of &mm_struct.mm_count vs &mm_struct.mm_users.
  96  */
  97 static inline void mmget(struct mm_struct *mm)
  98 {
  99         atomic_inc(&mm->mm_users);
 100 }
 101
 102 static inline bool mmget_not_zero(struct mm_struct *mm)
 103 {
 104         return atomic_inc_not_zero(&mm->mm_users);
 105 }
 106
 107 /* mmput gets rid of the mappings and all user-space */
 108 extern void mmput(struct mm_struct *);
 109 #ifdef CONFIG_MMU
 110 /* same as above but performs the slow path from the async context. Can
 111  * be called from the atomic context as well
 112  */
 113 void mmput_async(struct mm_struct *);
 114 #endif
 115
 116 /* Grab a reference to a task's mm, if it is not already going away */
 117 extern struct mm_struct *get_task_mm(struct task_struct *task);
 118 /*
 119  * Grab a reference to a task's mm, if it is not already going away
 120  * and ptrace_may_access with the mode parameter passed to it
 121  * succeeds.
 122  */
 123 extern struct mm_struct *mm_access(struct task_struct *task, unsigned int mode);
 124 /* Remove the current tasks stale references to the old mm_struct */
 125 extern void 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
 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 extern unsigned long
 138 arch_get_unmapped_area(struct file *, unsigned long, unsigned long,
 139                        unsigned long, unsigned long);
 140 extern unsigned long
 141 arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
 142                           unsigned long len, unsigned long pgoff,
 143                           unsigned long flags);
 144 #else
 145 static inline void arch_pick_mmap_layout(struct mm_struct *mm) {}
 146 #endif
 147
 148 static inline bool in_vfork(struct task_struct *tsk)
 149 {
 150         bool ret;
 151
 152         /*
 153          * need RCU to access ->real_parent if CLONE_VM was used along with
 154          * CLONE_PARENT.
 155          *
 156          * We check real_parent->mm == tsk->mm because CLONE_VFORK does not
 157          * imply CLONE_VM
 158          *
 159          * CLONE_VFORK can be used with CLONE_PARENT/CLONE_THREAD and thus
 160          * ->real_parent is not necessarily the task doing vfork(), so in
 161          * theory we can't rely on task_lock() if we want to dereference it.
 162          *
 163          * And in this case we can't trust the real_parent->mm == tsk->mm
 164          * check, it can be false negative. But we do not care, if init or
 165          * another oom-unkillable task does this it should blame itself.
 166          */
 167         rcu_read_lock();
 168         ret = tsk->vfork_done && tsk->real_parent->mm == tsk->mm;
 169         rcu_read_unlock();
 170
 171         return ret;
 172 }
 173
 174 /*
 175  * Applies per-task gfp context to the given allocation flags.
 176  * PF_MEMALLOC_NOIO implies GFP_NOIO
 177  * PF_MEMALLOC_NOFS implies GFP_NOFS
 178  */
 179 static inline gfp_t current_gfp_context(gfp_t flags)
 180 {
 181         /*
 182          * NOIO implies both NOIO and NOFS and it is a weaker context
 183          * so always make sure it makes precendence
 184          */
 185         if (unlikely(current->flags & PF_MEMALLOC_NOIO))
 186                 flags &= ~(__GFP_IO | __GFP_FS);
 187         else if (unlikely(current->flags & PF_MEMALLOC_NOFS))
 188                 flags &= ~__GFP_FS;
 189         return flags;
 190 }
 191
 192 #ifdef CONFIG_LOCKDEP
 193 extern void fs_reclaim_acquire(gfp_t gfp_mask);
 194 extern void fs_reclaim_release(gfp_t gfp_mask);
 195 #else
 196 static inline void fs_reclaim_acquire(gfp_t gfp_mask) { }
 197 static inline void fs_reclaim_release(gfp_t gfp_mask) { }
 198 #endif
 199
 200 static inline unsigned int memalloc_noio_save(void)
 201 {
 202         unsigned int flags = current->flags & PF_MEMALLOC_NOIO;
 203         current->flags |= PF_MEMALLOC_NOIO;
 204         return flags;
 205 }
 206
 207 static inline void memalloc_noio_restore(unsigned int flags)
 208 {
 209         current->flags = (current->flags & ~PF_MEMALLOC_NOIO) | flags;
 210 }
 211
 212 static inline unsigned int memalloc_nofs_save(void)
 213 {
 214         unsigned int flags = current->flags & PF_MEMALLOC_NOFS;
 215         current->flags |= PF_MEMALLOC_NOFS;
 216         return flags;
 217 }
 218
 219 static inline void memalloc_nofs_restore(unsigned int flags)
 220 {
 221         current->flags = (current->flags & ~PF_MEMALLOC_NOFS) | flags;
 222 }
 223
 224 static inline unsigned int memalloc_noreclaim_save(void)
 225 {
 226         unsigned int flags = current->flags & PF_MEMALLOC;
 227         current->flags |= PF_MEMALLOC;
 228         return flags;
 229 }
 230
 231 static inline void memalloc_noreclaim_restore(unsigned int flags)
 232 {
 233         current->flags = (current->flags & ~PF_MEMALLOC) | flags;
 234 }
 235
 236 #ifdef CONFIG_MEMBARRIER
 237 enum {
 238         MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY        = (1U << 0),
 239         MEMBARRIER_STATE_SWITCH_MM                      = (1U << 1),
 240 };
 241
 242 static inline void membarrier_execve(struct task_struct *t)
 243 {
 244         atomic_set(&t->mm->membarrier_state, 0);
 245 }
 246 #else
 247 static inline void membarrier_execve(struct task_struct *t)
 248 {
 249 }
 250 #endif
 251
 252 #endif /* _LINUX_SCHED_MM_H */