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 #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 static inline void mmgrab(struct mm_struct *mm)
  35 {
  36         atomic_inc(&mm->mm_count);
  37 }
  38
  39 extern void __mmdrop(struct mm_struct *mm);
  40
  41 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         if (unlikely(atomic_dec_and_test(&mm->mm_count)))
  49                 __mmdrop(mm);
  50 }
  51
  52 /*
  53  * This has to be called after a get_task_mm()/mmget_not_zero()
  54  * followed by taking the mmap_sem for writing before modifying the
  55  * vmas or anything the coredump pretends not to change from under it.
  56  *
  57  * NOTE: find_extend_vma() called from GUP context is the only place
  58  * that can modify the "mm" (notably the vm_start/end) under mmap_sem
  59  * for reading and outside the context of the process, so it is also
  60  * the only case that holds the mmap_sem for reading that must call
  61  * this function. Generally if the mmap_sem is hold for reading
  62  * there's no need of this check after get_task_mm()/mmget_not_zero().
  63  *
  64  * This function can be obsoleted and the check can be removed, after
  65  * the coredump code will hold the mmap_sem for writing before
  66  * invoking the ->core_dump methods.
  67  */
  68 static inline bool mmget_still_valid(struct mm_struct *mm)
  69 {
  70         return likely(!mm->core_state);
  71 }
  72
  73 /**
  74  * mmget() - Pin the address space associated with a &struct mm_struct.
  75  * @mm: The address space to pin.
  76  *
  77  * Make sure that the address space of the given &struct mm_struct doesn't
  78  * go away. This does not protect against parts of the address space being
  79  * modified or freed, however.
  80  *
  81  * Never use this function to pin this address space for an
  82  * unbounded/indefinite amount of time.
  83  *
  84  * Use mmput() to release the reference acquired by mmget().
  85  *
  86  * See also <Documentation/vm/active_mm.rst> for an in-depth explanation
  87  * of &mm_struct.mm_count vs &mm_struct.mm_users.
  88  */
  89 static inline void mmget(struct mm_struct *mm)
  90 {
  91         atomic_inc(&mm->mm_users);
  92 }
  93
  94 static inline bool mmget_not_zero(struct mm_struct *mm)
  95 {
  96         return atomic_inc_not_zero(&mm->mm_users);
  97 }
  98
  99 /* mmput gets rid of the mappings and all user-space */
 100 extern void mmput(struct mm_struct *);
 101 #ifdef CONFIG_MMU
 102 /* same as above but performs the slow path from the async context. Can
 103  * be called from the atomic context as well
 104  */
 105 void mmput_async(struct mm_struct *);
 106 #endif
 107
 108 /* Grab a reference to a task's mm, if it is not already going away */
 109 extern struct mm_struct *get_task_mm(struct task_struct *task);
 110 /*
 111  * Grab a reference to a task's mm, if it is not already going away
 112  * and ptrace_may_access with the mode parameter passed to it
 113  * succeeds.
 114  */
 115 extern struct mm_struct *mm_access(struct task_struct *task, unsigned int mode);
 116 /* Remove the current tasks stale references to the old mm_struct */
 117 extern void mm_release(struct task_struct *, struct mm_struct *);
 118
 119 #ifdef CONFIG_MEMCG
 120 extern void mm_update_next_owner(struct mm_struct *mm);
 121 #else
 122 static inline void mm_update_next_owner(struct mm_struct *mm)
 123 {
 124 }
 125 #endif /* CONFIG_MEMCG */
 126
 127 #ifdef CONFIG_MMU
 128 extern void arch_pick_mmap_layout(struct mm_struct *mm,
 129                                   struct rlimit *rlim_stack);
 130 extern unsigned long
 131 arch_get_unmapped_area(struct file *, unsigned long, unsigned long,
 132                        unsigned long, unsigned long);
 133 extern unsigned long
 134 arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
 135                           unsigned long len, unsigned long pgoff,
 136                           unsigned long flags);
 137 #else
 138 static inline void arch_pick_mmap_layout(struct mm_struct *mm,
 139                                          struct rlimit *rlim_stack) {}
 140 #endif
 141
 142 static inline bool in_vfork(struct task_struct *tsk)
 143 {
 144         bool ret;
 145
 146         /*
 147          * need RCU to access ->real_parent if CLONE_VM was used along with
 148          * CLONE_PARENT.
 149          *
 150          * We check real_parent->mm == tsk->mm because CLONE_VFORK does not
 151          * imply CLONE_VM
 152          *
 153          * CLONE_VFORK can be used with CLONE_PARENT/CLONE_THREAD and thus
 154          * ->real_parent is not necessarily the task doing vfork(), so in
 155          * theory we can't rely on task_lock() if we want to dereference it.
 156          *
 157          * And in this case we can't trust the real_parent->mm == tsk->mm
 158          * check, it can be false negative. But we do not care, if init or
 159          * another oom-unkillable task does this it should blame itself.
 160          */
 161         rcu_read_lock();
 162         ret = tsk->vfork_done && tsk->real_parent->mm == tsk->mm;
 163         rcu_read_unlock();
 164
 165         return ret;
 166 }
 167
 168 /*
 169  * Applies per-task gfp context to the given allocation flags.
 170  * PF_MEMALLOC_NOIO implies GFP_NOIO
 171  * PF_MEMALLOC_NOFS implies GFP_NOFS
 172  * PF_MEMALLOC_NOCMA implies no allocation from CMA region.
 173  */
 174 static inline gfp_t current_gfp_context(gfp_t flags)
 175 {
 176         if (unlikely(current->flags &
 177                      (PF_MEMALLOC_NOIO | PF_MEMALLOC_NOFS | PF_MEMALLOC_NOCMA))) {
 178                 /*
 179                  * NOIO implies both NOIO and NOFS and it is a weaker context
 180                  * so always make sure it makes precedence
 181                  */
 182                 if (current->flags & PF_MEMALLOC_NOIO)
 183                         flags &= ~(__GFP_IO | __GFP_FS);
 184                 else if (current->flags & PF_MEMALLOC_NOFS)
 185                         flags &= ~__GFP_FS;
 186 #ifdef CONFIG_CMA
 187                 if (current->flags & PF_MEMALLOC_NOCMA)
 188                         flags &= ~__GFP_MOVABLE;
 189 #endif
 190         }
 191         return flags;
 192 }
 193
 194 #ifdef CONFIG_LOCKDEP
 195 extern void __fs_reclaim_acquire(void);
 196 extern void __fs_reclaim_release(void);
 197 extern void fs_reclaim_acquire(gfp_t gfp_mask);
 198 extern void fs_reclaim_release(gfp_t gfp_mask);
 199 #else
 200 static inline void __fs_reclaim_acquire(void) { }
 201 static inline void __fs_reclaim_release(void) { }
 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 /**
 207  * memalloc_noio_save - Marks implicit GFP_NOIO allocation scope.
 208  *
 209  * This functions marks the beginning of the GFP_NOIO allocation scope.
 210  * All further allocations will implicitly drop __GFP_IO flag and so
 211  * they are safe for the IO critical section from the allocation recursion
 212  * point of view. Use memalloc_noio_restore to end the scope with flags
 213  * returned by this function.
 214  *
 215  * This function is safe to be used from any context.
 216  */
 217 static inline unsigned int memalloc_noio_save(void)
 218 {
 219         unsigned int flags = current->flags & PF_MEMALLOC_NOIO;
 220         current->flags |= PF_MEMALLOC_NOIO;
 221         return flags;
 222 }
 223
 224 /**
 225  * memalloc_noio_restore - Ends the implicit GFP_NOIO scope.
 226  * @flags: Flags to restore.
 227  *
 228  * Ends the implicit GFP_NOIO scope started by memalloc_noio_save function.
 229  * Always make sure that that the given flags is the return value from the
 230  * pairing memalloc_noio_save call.
 231  */
 232 static inline void memalloc_noio_restore(unsigned int flags)
 233 {
 234         current->flags = (current->flags & ~PF_MEMALLOC_NOIO) | flags;
 235 }
 236
 237 /**
 238  * memalloc_nofs_save - Marks implicit GFP_NOFS allocation scope.
 239  *
 240  * This functions marks the beginning of the GFP_NOFS allocation scope.
 241  * All further allocations will implicitly drop __GFP_FS flag and so
 242  * they are safe for the FS critical section from the allocation recursion
 243  * point of view. Use memalloc_nofs_restore to end the scope with flags
 244  * returned by this function.
 245  *
 246  * This function is safe to be used from any context.
 247  */
 248 static inline unsigned int memalloc_nofs_save(void)
 249 {
 250         unsigned int flags = current->flags & PF_MEMALLOC_NOFS;
 251         current->flags |= PF_MEMALLOC_NOFS;
 252         return flags;
 253 }
 254
 255 /**
 256  * memalloc_nofs_restore - Ends the implicit GFP_NOFS scope.
 257  * @flags: Flags to restore.
 258  *
 259  * Ends the implicit GFP_NOFS scope started by memalloc_nofs_save function.
 260  * Always make sure that that the given flags is the return value from the
 261  * pairing memalloc_nofs_save call.
 262  */
 263 static inline void memalloc_nofs_restore(unsigned int flags)
 264 {
 265         current->flags = (current->flags & ~PF_MEMALLOC_NOFS) | flags;
 266 }
 267
 268 static inline unsigned int memalloc_noreclaim_save(void)
 269 {
 270         unsigned int flags = current->flags & PF_MEMALLOC;
 271         current->flags |= PF_MEMALLOC;
 272         return flags;
 273 }
 274
 275 static inline void memalloc_noreclaim_restore(unsigned int flags)
 276 {
 277         current->flags = (current->flags & ~PF_MEMALLOC) | flags;
 278 }
 279
 280 #ifdef CONFIG_CMA
 281 static inline unsigned int memalloc_nocma_save(void)
 282 {
 283         unsigned int flags = current->flags & PF_MEMALLOC_NOCMA;
 284
 285         current->flags |= PF_MEMALLOC_NOCMA;
 286         return flags;
 287 }
 288
 289 static inline void memalloc_nocma_restore(unsigned int flags)
 290 {
 291         current->flags = (current->flags & ~PF_MEMALLOC_NOCMA) | flags;
 292 }
 293 #else
 294 static inline unsigned int memalloc_nocma_save(void)
 295 {
 296         return 0;
 297 }
 298
 299 static inline void memalloc_nocma_restore(unsigned int flags)
 300 {
 301 }
 302 #endif
 303
 304 #ifdef CONFIG_MEMCG
 305 /**
 306  * memalloc_use_memcg - Starts the remote memcg charging scope.
 307  * @memcg: memcg to charge.
 308  *
 309  * This function marks the beginning of the remote memcg charging scope. All the
 310  * __GFP_ACCOUNT allocations till the end of the scope will be charged to the
 311  * given memcg.
 312  *
 313  * NOTE: This function is not nesting safe.
 314  */
 315 static inline void memalloc_use_memcg(struct mem_cgroup *memcg)
 316 {
 317         WARN_ON_ONCE(current->active_memcg);
 318         current->active_memcg = memcg;
 319 }
 320
 321 /**
 322  * memalloc_unuse_memcg - Ends the remote memcg charging scope.
 323  *
 324  * This function marks the end of the remote memcg charging scope started by
 325  * memalloc_use_memcg().
 326  */
 327 static inline void memalloc_unuse_memcg(void)
 328 {
 329         current->active_memcg = NULL;
 330 }
 331 #else
 332 static inline void memalloc_use_memcg(struct mem_cgroup *memcg)
 333 {
 334 }
 335
 336 static inline void memalloc_unuse_memcg(void)
 337 {
 338 }
 339 #endif
 340
 341 #ifdef CONFIG_MEMBARRIER
 342 enum {
 343         MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY                = (1U << 0),
 344         MEMBARRIER_STATE_PRIVATE_EXPEDITED                      = (1U << 1),
 345         MEMBARRIER_STATE_GLOBAL_EXPEDITED_READY                 = (1U << 2),
 346         MEMBARRIER_STATE_GLOBAL_EXPEDITED                       = (1U << 3),
 347         MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE_READY      = (1U << 4),
 348         MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE            = (1U << 5),
 349 };
 350
 351 enum {
 352         MEMBARRIER_FLAG_SYNC_CORE       = (1U << 0),
 353 };
 354
 355 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
 356 #include <asm/membarrier.h>
 357 #endif
 358
 359 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
 360 {
 361         if (likely(!(atomic_read(&mm->membarrier_state) &
 362                      MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE)))
 363                 return;
 364         sync_core_before_usermode();
 365 }
 366
 367 static inline void membarrier_execve(struct task_struct *t)
 368 {
 369         atomic_set(&t->mm->membarrier_state, 0);
 370 }
 371 #else
 372 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
 373 static inline void membarrier_arch_switch_mm(struct mm_struct *prev,
 374                                              struct mm_struct *next,
 375                                              struct task_struct *tsk)
 376 {
 377 }
 378 #endif
 379 static inline void membarrier_execve(struct task_struct *t)
 380 {
 381 }
 382 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
 383 {
 384 }
 385 #endif
 386
 387 #endif /* _LINUX_SCHED_MM_H */