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 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        6530 : static inline void mmgrab(struct mm_struct *mm)
      35             : {
      36        6530 :         atomic_inc(&mm->mm_count);
      37        6530 : }
      38             : 
      39             : extern void __mmdrop(struct mm_struct *mm);
      40             : 
      41        9414 : 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       18829 :         if (unlikely(atomic_dec_and_test(&mm->mm_count)))
      49        2057 :                 __mmdrop(mm);
      50        9415 : }
      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             :  */
      68         322 : static inline void mmget(struct mm_struct *mm)
      69             : {
      70         322 :         atomic_inc(&mm->mm_users);
      71         322 : }
      72             : 
      73          92 : static inline bool mmget_not_zero(struct mm_struct *mm)
      74             : {
      75          92 :         return atomic_inc_not_zero(&mm->mm_users);
      76             : }
      77             : 
      78             : /* mmput gets rid of the mappings and all user-space */
      79             : extern 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             :  */
      84             : void mmput_async(struct mm_struct *);
      85             : #endif
      86             : 
      87             : /* Grab a reference to a task's mm, if it is not already going away */
      88             : extern 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             :  */
      94             : extern struct mm_struct *mm_access(struct task_struct *task, unsigned int mode);
      95             : /* Remove the current tasks stale references to the old mm_struct on exit() */
      96             : extern void exit_mm_release(struct task_struct *, struct mm_struct *);
      97             : /* Remove the current tasks stale references to the old mm_struct on exec() */
      98             : extern void exec_mm_release(struct task_struct *, struct mm_struct *);
      99             : 
     100             : #ifdef CONFIG_MEMCG
     101             : extern void mm_update_next_owner(struct mm_struct *mm);
     102             : #else
     103        1020 : static inline void mm_update_next_owner(struct mm_struct *mm)
     104             : {
     105        1020 : }
     106             : #endif /* CONFIG_MEMCG */
     107             : 
     108             : #ifdef CONFIG_MMU
     109             : extern void arch_pick_mmap_layout(struct mm_struct *mm,
     110             :                                   struct rlimit *rlim_stack);
     111             : extern unsigned long
     112             : arch_get_unmapped_area(struct file *, unsigned long, unsigned long,
     113             :                        unsigned long, unsigned long);
     114             : extern unsigned long
     115             : arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
     116             :                           unsigned long len, unsigned long pgoff,
     117             :                           unsigned long flags);
     118             : #else
     119             : static inline void arch_pick_mmap_layout(struct mm_struct *mm,
     120             :                                          struct rlimit *rlim_stack) {}
     121             : #endif
     122             : 
     123           0 : static inline bool in_vfork(struct task_struct *tsk)
     124             : {
     125           0 :         bool ret;
     126             : 
     127             :         /*
     128             :          * need RCU to access ->real_parent if CLONE_VM was used along with
     129             :          * CLONE_PARENT.
     130             :          *
     131             :          * We check real_parent->mm == tsk->mm because CLONE_VFORK does not
     132             :          * imply CLONE_VM
     133             :          *
     134             :          * CLONE_VFORK can be used with CLONE_PARENT/CLONE_THREAD and thus
     135             :          * ->real_parent is not necessarily the task doing vfork(), so in
     136             :          * theory we can't rely on task_lock() if we want to dereference it.
     137             :          *
     138             :          * And in this case we can't trust the real_parent->mm == tsk->mm
     139             :          * check, it can be false negative. But we do not care, if init or
     140             :          * another oom-unkillable task does this it should blame itself.
     141             :          */
     142           0 :         rcu_read_lock();
     143           0 :         ret = tsk->vfork_done &&
     144           0 :                         rcu_dereference(tsk->real_parent)->mm == tsk->mm;
     145           0 :         rcu_read_unlock();
     146             : 
     147           0 :         return ret;
     148             : }
     149             : 
     150             : /*
     151             :  * Applies per-task gfp context to the given allocation flags.
     152             :  * PF_MEMALLOC_NOIO implies GFP_NOIO
     153             :  * PF_MEMALLOC_NOFS implies GFP_NOFS
     154             :  */
     155     3298088 : static inline gfp_t current_gfp_context(gfp_t flags)
     156             : {
     157     3298088 :         unsigned int pflags = READ_ONCE(current->flags);
     158             : 
     159     3298088 :         if (unlikely(pflags & (PF_MEMALLOC_NOIO | PF_MEMALLOC_NOFS))) {
     160             :                 /*
     161             :                  * NOIO implies both NOIO and NOFS and it is a weaker context
     162             :                  * so always make sure it makes precedence
     163             :                  */
     164      114094 :                 if (pflags & PF_MEMALLOC_NOIO)
     165           0 :                         flags &= ~(__GFP_IO | __GFP_FS);
     166      114094 :                 else if (pflags & PF_MEMALLOC_NOFS)
     167      114094 :                         flags &= ~__GFP_FS;
     168             :         }
     169     3298088 :         return flags;
     170             : }
     171             : 
     172             : #ifdef CONFIG_LOCKDEP
     173             : extern void __fs_reclaim_acquire(void);
     174             : extern void __fs_reclaim_release(void);
     175             : extern void fs_reclaim_acquire(gfp_t gfp_mask);
     176             : extern void fs_reclaim_release(gfp_t gfp_mask);
     177             : #else
     178             : static inline void __fs_reclaim_acquire(void) { }
     179             : static inline void __fs_reclaim_release(void) { }
     180             : static inline void fs_reclaim_acquire(gfp_t gfp_mask) { }
     181             : static inline void fs_reclaim_release(gfp_t gfp_mask) { }
     182             : #endif
     183             : 
     184             : /**
     185             :  * might_alloc - Mark possible allocation sites
     186             :  * @gfp_mask: gfp_t flags that would be used to allocate
     187             :  *
     188             :  * Similar to might_sleep() and other annotations, this can be used in functions
     189             :  * that might allocate, but often don't. Compiles to nothing without
     190             :  * CONFIG_LOCKDEP. Includes a conditional might_sleep() if @gfp allows blocking.
     191             :  */
     192     1471217 : static inline void might_alloc(gfp_t gfp_mask)
     193             : {
     194     1471217 :         fs_reclaim_acquire(gfp_mask);
     195     1470977 :         fs_reclaim_release(gfp_mask);
     196             : 
     197     1470838 :         might_sleep_if(gfpflags_allow_blocking(gfp_mask));
     198     1471380 : }
     199             : 
     200             : /**
     201             :  * memalloc_noio_save - Marks implicit GFP_NOIO allocation scope.
     202             :  *
     203             :  * This functions marks the beginning of the GFP_NOIO allocation scope.
     204             :  * All further allocations will implicitly drop __GFP_IO flag and so
     205             :  * they are safe for the IO critical section from the allocation recursion
     206             :  * point of view. Use memalloc_noio_restore to end the scope with flags
     207             :  * returned by this function.
     208             :  *
     209             :  * This function is safe to be used from any context.
     210             :  */
     211           0 : static inline unsigned int memalloc_noio_save(void)
     212             : {
     213           0 :         unsigned int flags = current->flags & PF_MEMALLOC_NOIO;
     214           0 :         current->flags |= PF_MEMALLOC_NOIO;
     215           0 :         return flags;
     216             : }
     217             : 
     218             : /**
     219             :  * memalloc_noio_restore - Ends the implicit GFP_NOIO scope.
     220             :  * @flags: Flags to restore.
     221             :  *
     222             :  * Ends the implicit GFP_NOIO scope started by memalloc_noio_save function.
     223             :  * Always make sure that the given flags is the return value from the
     224             :  * pairing memalloc_noio_save call.
     225             :  */
     226           0 : static inline void memalloc_noio_restore(unsigned int flags)
     227             : {
     228           0 :         current->flags = (current->flags & ~PF_MEMALLOC_NOIO) | flags;
     229             : }
     230             : 
     231             : /**
     232             :  * memalloc_nofs_save - Marks implicit GFP_NOFS allocation scope.
     233             :  *
     234             :  * This functions marks the beginning of the GFP_NOFS allocation scope.
     235             :  * All further allocations will implicitly drop __GFP_FS flag and so
     236             :  * they are safe for the FS critical section from the allocation recursion
     237             :  * point of view. Use memalloc_nofs_restore to end the scope with flags
     238             :  * returned by this function.
     239             :  *
     240             :  * This function is safe to be used from any context.
     241             :  */
     242        7912 : static inline unsigned int memalloc_nofs_save(void)
     243             : {
     244        7912 :         unsigned int flags = current->flags & PF_MEMALLOC_NOFS;
     245        7912 :         current->flags |= PF_MEMALLOC_NOFS;
     246        7912 :         return flags;
     247             : }
     248             : 
     249             : /**
     250             :  * memalloc_nofs_restore - Ends the implicit GFP_NOFS scope.
     251             :  * @flags: Flags to restore.
     252             :  *
     253             :  * Ends the implicit GFP_NOFS scope started by memalloc_nofs_save function.
     254             :  * Always make sure that the given flags is the return value from the
     255             :  * pairing memalloc_nofs_save call.
     256             :  */
     257        7911 : static inline void memalloc_nofs_restore(unsigned int flags)
     258             : {
     259        1656 :         current->flags = (current->flags & ~PF_MEMALLOC_NOFS) | flags;
     260             : }
     261             : 
     262           0 : static inline unsigned int memalloc_noreclaim_save(void)
     263             : {
     264           0 :         unsigned int flags = current->flags & PF_MEMALLOC;
     265           0 :         current->flags |= PF_MEMALLOC;
     266           0 :         return flags;
     267             : }
     268             : 
     269           0 : static inline void memalloc_noreclaim_restore(unsigned int flags)
     270             : {
     271           0 :         current->flags = (current->flags & ~PF_MEMALLOC) | flags;
     272           0 : }
     273             : 
     274             : #ifdef CONFIG_CMA
     275             : static inline unsigned int memalloc_nocma_save(void)
     276             : {
     277             :         unsigned int flags = current->flags & PF_MEMALLOC_NOCMA;
     278             : 
     279             :         current->flags |= PF_MEMALLOC_NOCMA;
     280             :         return flags;
     281             : }
     282             : 
     283             : static inline void memalloc_nocma_restore(unsigned int flags)
     284             : {
     285             :         current->flags = (current->flags & ~PF_MEMALLOC_NOCMA) | flags;
     286             : }
     287             : #else
     288             : static inline unsigned int memalloc_nocma_save(void)
     289             : {
     290             :         return 0;
     291             : }
     292             : 
     293             : static inline void memalloc_nocma_restore(unsigned int flags)
     294             : {
     295             : }
     296             : #endif
     297             : 
     298             : #ifdef CONFIG_MEMCG
     299             : DECLARE_PER_CPU(struct mem_cgroup *, int_active_memcg);
     300             : /**
     301             :  * set_active_memcg - Starts the remote memcg charging scope.
     302             :  * @memcg: memcg to charge.
     303             :  *
     304             :  * This function marks the beginning of the remote memcg charging scope. All the
     305             :  * __GFP_ACCOUNT allocations till the end of the scope will be charged to the
     306             :  * given memcg.
     307             :  *
     308             :  * NOTE: This function can nest. Users must save the return value and
     309             :  * reset the previous value after their own charging scope is over.
     310             :  */
     311             : static inline struct mem_cgroup *
     312             : set_active_memcg(struct mem_cgroup *memcg)
     313             : {
     314             :         struct mem_cgroup *old;
     315             : 
     316             :         if (in_interrupt()) {
     317             :                 old = this_cpu_read(int_active_memcg);
     318             :                 this_cpu_write(int_active_memcg, memcg);
     319             :         } else {
     320             :                 old = current->active_memcg;
     321             :                 current->active_memcg = memcg;
     322             :         }
     323             : 
     324             :         return old;
     325             : }
     326             : #else
     327             : static inline struct mem_cgroup *
     328       11972 : set_active_memcg(struct mem_cgroup *memcg)
     329             : {
     330       11972 :         return NULL;
     331             : }
     332             : #endif
     333             : 
     334             : #ifdef CONFIG_MEMBARRIER
     335             : enum {
     336             :         MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY                = (1U << 0),
     337             :         MEMBARRIER_STATE_PRIVATE_EXPEDITED                      = (1U << 1),
     338             :         MEMBARRIER_STATE_GLOBAL_EXPEDITED_READY                 = (1U << 2),
     339             :         MEMBARRIER_STATE_GLOBAL_EXPEDITED                       = (1U << 3),
     340             :         MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE_READY      = (1U << 4),
     341             :         MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE            = (1U << 5),
     342             :         MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ_READY           = (1U << 6),
     343             :         MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ                 = (1U << 7),
     344             : };
     345             : 
     346             : enum {
     347             :         MEMBARRIER_FLAG_SYNC_CORE       = (1U << 0),
     348             :         MEMBARRIER_FLAG_RSEQ            = (1U << 1),
     349             : };
     350             : 
     351             : #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
     352             : #include <asm/membarrier.h>
     353             : #endif
     354             : 
     355             : static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
     356             : {
     357             :         if (current->mm != mm)
     358             :                 return;
     359             :         if (likely(!(atomic_read(&mm->membarrier_state) &
     360             :                      MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE)))
     361             :                 return;
     362             :         sync_core_before_usermode();
     363             : }
     364             : 
     365             : extern void membarrier_exec_mmap(struct mm_struct *mm);
     366             : 
     367             : extern void membarrier_update_current_mm(struct mm_struct *next_mm);
     368             : 
     369             : #else
     370             : #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
     371             : static inline void membarrier_arch_switch_mm(struct mm_struct *prev,
     372             :                                              struct mm_struct *next,
     373             :                                              struct task_struct *tsk)
     374             : {
     375             : }
     376             : #endif
     377        1021 : static inline void membarrier_exec_mmap(struct mm_struct *mm)
     378             : {
     379        1021 : }
     380        7345 : static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
     381             : {
     382        7345 : }
     383           0 : static inline void membarrier_update_current_mm(struct mm_struct *next_mm)
     384             : {
     385           0 : }
     386             : #endif
     387             : 
     388             : #endif /* _LINUX_SCHED_MM_H */