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  * It also has to be called when mmgrab() is used in the context of
  66  * the process, but then the mm_count refcount is transferred outside
  67  * the context of the process to run down_write() on that pinned mm.
  68  *
  69  * NOTE: find_extend_vma() called from GUP context is the only place
  70  * that can modify the "mm" (notably the vm_start/end) under mmap_sem
  71  * for reading and outside the context of the process, so it is also
  72  * the only case that holds the mmap_sem for reading that must call
  73  * this function. Generally if the mmap_sem is hold for reading
  74  * there's no need of this check after get_task_mm()/mmget_not_zero().
  75  *
  76  * This function can be obsoleted and the check can be removed, after
  77  * the coredump code will hold the mmap_sem for writing before
  78  * invoking the ->core_dump methods.
  79  */
  80 static inline bool mmget_still_valid(struct mm_struct *mm)
  81 {
  82         return likely(!mm->core_state);
  83 }
  84
  85 /**
  86  * mmget() - Pin the address space associated with a &struct mm_struct.
  87  * @mm: The address space to pin.
  88  *
  89  * Make sure that the address space of the given &struct mm_struct doesn't
  90  * go away. This does not protect against parts of the address space being
  91  * modified or freed, however.
  92  *
  93  * Never use this function to pin this address space for an
  94  * unbounded/indefinite amount of time.
  95  *
  96  * Use mmput() to release the reference acquired by mmget().
  97  *
  98  * See also <Documentation/vm/active_mm.txt> for an in-depth explanation
  99  * of &mm_struct.mm_count vs &mm_struct.mm_users.
 100  */
 101 static inline void mmget(struct mm_struct *mm)
 102 {
 103         atomic_inc(&mm->mm_users);
 104 }
 105
 106 static inline bool mmget_not_zero(struct mm_struct *mm)
 107 {
 108         return atomic_inc_not_zero(&mm->mm_users);
 109 }
 110
 111 /* mmput gets rid of the mappings and all user-space */
 112 extern void mmput(struct mm_struct *);
 113 #ifdef CONFIG_MMU
 114 /* same as above but performs the slow path from the async context. Can
 115  * be called from the atomic context as well
 116  */
 117 void mmput_async(struct mm_struct *);
 118 #endif
 119
 120 /* Grab a reference to a task's mm, if it is not already going away */
 121 extern struct mm_struct *get_task_mm(struct task_struct *task);
 122 /*
 123  * Grab a reference to a task's mm, if it is not already going away
 124  * and ptrace_may_access with the mode parameter passed to it
 125  * succeeds.
 126  */
 127 extern struct mm_struct *mm_access(struct task_struct *task, unsigned int mode);
 128 /* Remove the current tasks stale references to the old mm_struct on exit() */
 129 extern void exit_mm_release(struct task_struct *, struct mm_struct *);
 130 /* Remove the current tasks stale references to the old mm_struct on exec() */
 131 extern void exec_mm_release(struct task_struct *, struct mm_struct *);
 132
 133 #ifdef CONFIG_MEMCG
 134 extern void mm_update_next_owner(struct mm_struct *mm);
 135 #else
 136 static inline void mm_update_next_owner(struct mm_struct *mm)
 137 {
 138 }
 139 #endif /* CONFIG_MEMCG */
 140
 141 #ifdef CONFIG_MMU
 142 extern void arch_pick_mmap_layout(struct mm_struct *mm);
 143 extern unsigned long
 144 arch_get_unmapped_area(struct file *, unsigned long, unsigned long,
 145                        unsigned long, unsigned long);
 146 extern unsigned long
 147 arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
 148                           unsigned long len, unsigned long pgoff,
 149                           unsigned long flags);
 150 #else
 151 static inline void arch_pick_mmap_layout(struct mm_struct *mm) {}
 152 #endif
 153
 154 static inline bool in_vfork(struct task_struct *tsk)
 155 {
 156         bool ret;
 157
 158         /*
 159          * need RCU to access ->real_parent if CLONE_VM was used along with
 160          * CLONE_PARENT.
 161          *
 162          * We check real_parent->mm == tsk->mm because CLONE_VFORK does not
 163          * imply CLONE_VM
 164          *
 165          * CLONE_VFORK can be used with CLONE_PARENT/CLONE_THREAD and thus
 166          * ->real_parent is not necessarily the task doing vfork(), so in
 167          * theory we can't rely on task_lock() if we want to dereference it.
 168          *
 169          * And in this case we can't trust the real_parent->mm == tsk->mm
 170          * check, it can be false negative. But we do not care, if init or
 171          * another oom-unkillable task does this it should blame itself.
 172          */
 173         rcu_read_lock();
 174         ret = tsk->vfork_done && tsk->real_parent->mm == tsk->mm;
 175         rcu_read_unlock();
 176
 177         return ret;
 178 }
 179
 180 /*
 181  * Applies per-task gfp context to the given allocation flags.
 182  * PF_MEMALLOC_NOIO implies GFP_NOIO
 183  * PF_MEMALLOC_NOFS implies GFP_NOFS
 184  */
 185 static inline gfp_t current_gfp_context(gfp_t flags)
 186 {
 187         /*
 188          * NOIO implies both NOIO and NOFS and it is a weaker context
 189          * so always make sure it makes precendence
 190          */
 191         if (unlikely(current->flags & PF_MEMALLOC_NOIO))
 192                 flags &= ~(__GFP_IO | __GFP_FS);
 193         else if (unlikely(current->flags & PF_MEMALLOC_NOFS))
 194                 flags &= ~__GFP_FS;
 195         return flags;
 196 }
 197
 198 #ifdef CONFIG_LOCKDEP
 199 extern void fs_reclaim_acquire(gfp_t gfp_mask);
 200 extern void fs_reclaim_release(gfp_t gfp_mask);
 201 #else
 202 static inline void fs_reclaim_acquire(gfp_t gfp_mask) { }
 203 static inline void fs_reclaim_release(gfp_t gfp_mask) { }
 204 #endif
 205
 206 static inline unsigned int memalloc_noio_save(void)
 207 {
 208         unsigned int flags = current->flags & PF_MEMALLOC_NOIO;
 209         current->flags |= PF_MEMALLOC_NOIO;
 210         return flags;
 211 }
 212
 213 static inline void memalloc_noio_restore(unsigned int flags)
 214 {
 215         current->flags = (current->flags & ~PF_MEMALLOC_NOIO) | flags;
 216 }
 217
 218 static inline unsigned int memalloc_nofs_save(void)
 219 {
 220         unsigned int flags = current->flags & PF_MEMALLOC_NOFS;
 221         current->flags |= PF_MEMALLOC_NOFS;
 222         return flags;
 223 }
 224
 225 static inline void memalloc_nofs_restore(unsigned int flags)
 226 {
 227         current->flags = (current->flags & ~PF_MEMALLOC_NOFS) | flags;
 228 }
 229
 230 static inline unsigned int memalloc_noreclaim_save(void)
 231 {
 232         unsigned int flags = current->flags & PF_MEMALLOC;
 233         current->flags |= PF_MEMALLOC;
 234         return flags;
 235 }
 236
 237 static inline void memalloc_noreclaim_restore(unsigned int flags)
 238 {
 239         current->flags = (current->flags & ~PF_MEMALLOC) | flags;
 240 }
 241
 242 #ifdef CONFIG_MEMBARRIER
 243 enum {
 244         MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY        = (1U << 0),
 245         MEMBARRIER_STATE_SWITCH_MM                      = (1U << 1),
 246 };
 247
 248 static inline void membarrier_execve(struct task_struct *t)
 249 {
 250         atomic_set(&t->mm->membarrier_state, 0);
 251 }
 252 #else
 253 static inline void membarrier_execve(struct task_struct *t)
 254 {
 255 }
 256 #endif
 257
 258 #endif /* _LINUX_SCHED_MM_H */