Branch data     Line data    Source code

       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                 :            :  */
      34                 :    3059064 : static inline void mmgrab(struct mm_struct *mm)
      35                 :            : {
      36                 :    3239245 :         atomic_inc(&mm->mm_count);
      37                 :    3050048 : }
      38                 :            : 
      39                 :            : extern void __mmdrop(struct mm_struct *mm);
      40                 :            : 
      41                 :    3528537 : 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         [ +  + ]:    7058343 :         if (unlikely(atomic_dec_and_test(&mm->mm_count)))
      49                 :     290505 :                 __mmdrop(mm);
      50                 :    3529810 : }
      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                 :            :  */
      74                 :            : static inline bool mmget_still_valid(struct mm_struct *mm)
      75                 :            : {
      76                 :         33 :         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                 :            :  */
      95                 :            : static inline void mmget(struct mm_struct *mm)
      96                 :            : {
      97                 :     113590 :         atomic_inc(&mm->mm_users);
      98                 :            : }
      99                 :            : 
     100                 :            : static inline bool mmget_not_zero(struct mm_struct *mm)
     101                 :            : {
     102                 :      18977 :         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
     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
     146                 :            : static inline void arch_pick_mmap_layout(struct mm_struct *mm,
     147                 :            :                                          struct rlimit *rlim_stack) {}
     148                 :            : #endif
     149                 :            : 
     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   [ #  #  #  # ]:          0 :         ret = tsk->vfork_done && tsk->real_parent->mm == tsk->mm;
     171                 :            :         rcu_read_unlock();
     172                 :            : 
     173                 :            :         return ret;
     174                 :            : }
     175                 :            : 
     176                 :            : /*
     177                 :            :  * Applies per-task gfp context to the given allocation flags.
     178                 :            :  * PF_MEMALLOC_NOIO implies GFP_NOIO
     179                 :            :  * PF_MEMALLOC_NOFS implies GFP_NOFS
     180                 :            :  * PF_MEMALLOC_NOCMA implies no allocation from CMA region.
     181                 :            :  */
     182                 :       3726 : static inline gfp_t current_gfp_context(gfp_t flags)
     183                 :            : {
     184         [ -  + ]:       3726 :         if (unlikely(current->flags &
     185                 :            :                      (PF_MEMALLOC_NOIO | PF_MEMALLOC_NOFS | PF_MEMALLOC_NOCMA))) {
     186                 :            :                 /*
     187                 :            :                  * NOIO implies both NOIO and NOFS and it is a weaker context
     188                 :            :                  * so always make sure it makes precedence
     189                 :            :                  */
     190         [ #  # ]:          0 :                 if (current->flags & PF_MEMALLOC_NOIO)
     191                 :          0 :                         flags &= ~(__GFP_IO | __GFP_FS);
     192         [ #  # ]:          0 :                 else if (current->flags & PF_MEMALLOC_NOFS)
     193                 :          0 :                         flags &= ~__GFP_FS;
     194                 :            : #ifdef CONFIG_CMA
     195         [ #  # ]:          0 :                 if (current->flags & PF_MEMALLOC_NOCMA)
     196                 :          0 :                         flags &= ~__GFP_MOVABLE;
     197                 :            : #endif
     198                 :            :         }
     199                 :       3726 :         return flags;
     200                 :            : }
     201                 :            : 
     202                 :            : #ifdef CONFIG_LOCKDEP
     203                 :            : extern void __fs_reclaim_acquire(void);
     204                 :            : extern void __fs_reclaim_release(void);
     205                 :            : extern void fs_reclaim_acquire(gfp_t gfp_mask);
     206                 :            : extern void fs_reclaim_release(gfp_t gfp_mask);
     207                 :            : #else
     208                 :            : static inline void __fs_reclaim_acquire(void) { }
     209                 :            : static inline void __fs_reclaim_release(void) { }
     210                 :            : static inline void fs_reclaim_acquire(gfp_t gfp_mask) { }
     211                 :            : static inline void fs_reclaim_release(gfp_t gfp_mask) { }
     212                 :            : #endif
     213                 :            : 
     214                 :            : /**
     215                 :            :  * memalloc_noio_save - Marks implicit GFP_NOIO allocation scope.
     216                 :            :  *
     217                 :            :  * This functions marks the beginning of the GFP_NOIO allocation scope.
     218                 :            :  * All further allocations will implicitly drop __GFP_IO flag and so
     219                 :            :  * they are safe for the IO critical section from the allocation recursion
     220                 :            :  * point of view. Use memalloc_noio_restore to end the scope with flags
     221                 :            :  * returned by this function.
     222                 :            :  *
     223                 :            :  * This function is safe to be used from any context.
     224                 :            :  */
     225                 :            : static inline unsigned int memalloc_noio_save(void)
     226                 :            : {
     227                 :          0 :         unsigned int flags = current->flags & PF_MEMALLOC_NOIO;
     228                 :          0 :         current->flags |= PF_MEMALLOC_NOIO;
     229                 :            :         return flags;
     230                 :            : }
     231                 :            : 
     232                 :            : /**
     233                 :            :  * memalloc_noio_restore - Ends the implicit GFP_NOIO scope.
     234                 :            :  * @flags: Flags to restore.
     235                 :            :  *
     236                 :            :  * Ends the implicit GFP_NOIO scope started by memalloc_noio_save function.
     237                 :            :  * Always make sure that that the given flags is the return value from the
     238                 :            :  * pairing memalloc_noio_save call.
     239                 :            :  */
     240                 :            : static inline void memalloc_noio_restore(unsigned int flags)
     241                 :            : {
     242                 :          0 :         current->flags = (current->flags & ~PF_MEMALLOC_NOIO) | flags;
     243                 :            : }
     244                 :            : 
     245                 :            : /**
     246                 :            :  * memalloc_nofs_save - Marks implicit GFP_NOFS allocation scope.
     247                 :            :  *
     248                 :            :  * This functions marks the beginning of the GFP_NOFS allocation scope.
     249                 :            :  * All further allocations will implicitly drop __GFP_FS flag and so
     250                 :            :  * they are safe for the FS critical section from the allocation recursion
     251                 :            :  * point of view. Use memalloc_nofs_restore to end the scope with flags
     252                 :            :  * returned by this function.
     253                 :            :  *
     254                 :            :  * This function is safe to be used from any context.
     255                 :            :  */
     256                 :            : static inline unsigned int memalloc_nofs_save(void)
     257                 :            : {
     258                 :     791472 :         unsigned int flags = current->flags & PF_MEMALLOC_NOFS;
     259                 :     791472 :         current->flags |= PF_MEMALLOC_NOFS;
     260                 :            :         return flags;
     261                 :            : }
     262                 :            : 
     263                 :            : /**
     264                 :            :  * memalloc_nofs_restore - Ends the implicit GFP_NOFS scope.
     265                 :            :  * @flags: Flags to restore.
     266                 :            :  *
     267                 :            :  * Ends the implicit GFP_NOFS scope started by memalloc_nofs_save function.
     268                 :            :  * Always make sure that that the given flags is the return value from the
     269                 :            :  * pairing memalloc_nofs_save call.
     270                 :            :  */
     271                 :            : static inline void memalloc_nofs_restore(unsigned int flags)
     272                 :            : {
     273                 :     791286 :         current->flags = (current->flags & ~PF_MEMALLOC_NOFS) | flags;
     274                 :            : }
     275                 :            : 
     276                 :            : static inline unsigned int memalloc_noreclaim_save(void)
     277                 :            : {
     278                 :          0 :         unsigned int flags = current->flags & PF_MEMALLOC;
     279                 :          0 :         current->flags |= PF_MEMALLOC;
     280                 :            :         return flags;
     281                 :            : }
     282                 :            : 
     283                 :            : static inline void memalloc_noreclaim_restore(unsigned int flags)
     284                 :            : {
     285                 :          0 :         current->flags = (current->flags & ~PF_MEMALLOC) | flags;
     286                 :            : }
     287                 :            : 
     288                 :            : #ifdef CONFIG_CMA
     289                 :            : static inline unsigned int memalloc_nocma_save(void)
     290                 :            : {
     291                 :          0 :         unsigned int flags = current->flags & PF_MEMALLOC_NOCMA;
     292                 :            : 
     293                 :          0 :         current->flags |= PF_MEMALLOC_NOCMA;
     294                 :            :         return flags;
     295                 :            : }
     296                 :            : 
     297                 :            : static inline void memalloc_nocma_restore(unsigned int flags)
     298                 :            : {
     299                 :          0 :         current->flags = (current->flags & ~PF_MEMALLOC_NOCMA) | flags;
     300                 :            : }
     301                 :            : #else
     302                 :            : static inline unsigned int memalloc_nocma_save(void)
     303                 :            : {
     304                 :            :         return 0;
     305                 :            : }
     306                 :            : 
     307                 :            : static inline void memalloc_nocma_restore(unsigned int flags)
     308                 :            : {
     309                 :            : }
     310                 :            : #endif
     311                 :            : 
     312                 :            : #ifdef CONFIG_MEMCG
     313                 :            : /**
     314                 :            :  * memalloc_use_memcg - Starts the remote memcg charging scope.
     315                 :            :  * @memcg: memcg to charge.
     316                 :            :  *
     317                 :            :  * This function marks the beginning of the remote memcg charging scope. All the
     318                 :            :  * __GFP_ACCOUNT allocations till the end of the scope will be charged to the
     319                 :            :  * given memcg.
     320                 :            :  *
     321                 :            :  * NOTE: This function is not nesting safe.
     322                 :            :  */
     323                 :    1362221 : static inline void memalloc_use_memcg(struct mem_cgroup *memcg)
     324                 :            : {
     325   [ -  +  #  # ]:    1362221 :         WARN_ON_ONCE(current->active_memcg);
     326                 :    1362221 :         current->active_memcg = memcg;
     327                 :    1362221 : }
     328                 :            : 
     329                 :            : /**
     330                 :            :  * memalloc_unuse_memcg - Ends the remote memcg charging scope.
     331                 :            :  *
     332                 :            :  * This function marks the end of the remote memcg charging scope started by
     333                 :            :  * memalloc_use_memcg().
     334                 :            :  */
     335                 :            : static inline void memalloc_unuse_memcg(void)
     336                 :            : {
     337                 :    1362293 :         current->active_memcg = NULL;
     338                 :            : }
     339                 :            : #else
     340                 :            : static inline void memalloc_use_memcg(struct mem_cgroup *memcg)
     341                 :            : {
     342                 :            : }
     343                 :            : 
     344                 :            : static inline void memalloc_unuse_memcg(void)
     345                 :            : {
     346                 :            : }
     347                 :            : #endif
     348                 :            : 
     349                 :            : #ifdef CONFIG_MEMBARRIER
     350                 :            : enum {
     351                 :            :         MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY                = (1U << 0),
     352                 :            :         MEMBARRIER_STATE_PRIVATE_EXPEDITED                      = (1U << 1),
     353                 :            :         MEMBARRIER_STATE_GLOBAL_EXPEDITED_READY                 = (1U << 2),
     354                 :            :         MEMBARRIER_STATE_GLOBAL_EXPEDITED                       = (1U << 3),
     355                 :            :         MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE_READY      = (1U << 4),
     356                 :            :         MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE            = (1U << 5),
     357                 :            : };
     358                 :            : 
     359                 :            : enum {
     360                 :            :         MEMBARRIER_FLAG_SYNC_CORE       = (1U << 0),
     361                 :            : };
     362                 :            : 
     363                 :            : #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
     364                 :            : #include <asm/membarrier.h>
     365                 :            : #endif
     366                 :            : 
     367                 :            : static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
     368                 :            : {
     369         [ +  + ]:    3230035 :         if (current->mm != mm)
     370                 :            :                 return;
     371                 :            :         if (likely(!(atomic_read(&mm->membarrier_state) &
     372                 :            :                      MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE)))
     373                 :            :                 return;
     374                 :            :         sync_core_before_usermode();
     375                 :            : }
     376                 :            : 
     377                 :            : extern void membarrier_exec_mmap(struct mm_struct *mm);
     378                 :            : 
     379                 :            : #else
     380                 :            : #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
     381                 :            : static inline void membarrier_arch_switch_mm(struct mm_struct *prev,
     382                 :            :                                              struct mm_struct *next,
     383                 :            :                                              struct task_struct *tsk)
     384                 :            : {
     385                 :            : }
     386                 :            : #endif
     387                 :            : static inline void membarrier_exec_mmap(struct mm_struct *mm)
     388                 :            : {
     389                 :            : }
     390                 :            : static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
     391                 :            : {
     392                 :            : }
     393                 :            : #endif
     394                 :            : 
     395                 :            : #endif /* _LINUX_SCHED_MM_H */