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  * It also has to be called when mmgrab() is used in the context of
  58  * the process, but then the mm_count refcount is transferred outside
  59  * the context of the process to run down_write() on that pinned mm.
  60  *
  61  * NOTE: find_extend_vma() called from GUP context is the only place
  62  * that can modify the "mm" (notably the vm_start/end) under mmap_sem
  63  * for reading and outside the context of the process, so it is also
  64  * the only case that holds the mmap_sem for reading that must call
  65  * this function. Generally if the mmap_sem is hold for reading
  66  * there's no need of this check after get_task_mm()/mmget_not_zero().
  67  *
  68  * This function can be obsoleted and the check can be removed, after
  69  * the coredump code will hold the mmap_sem for writing before
  70  * invoking the ->core_dump methods.
  71  */
  72 static inline bool mmget_still_valid(struct mm_struct *mm)
  73 {
  74         return likely(!mm->core_state);
  75 }
  76
  77 /**
  78  * mmget() - Pin the address space associated with a &struct mm_struct.
  79  * @mm: The address space to pin.
  80  *
  81  * Make sure that the address space of the given &struct mm_struct doesn't
  82  * go away. This does not protect against parts of the address space being
  83  * modified or freed, however.
  84  *
  85  * Never use this function to pin this address space for an
  86  * unbounded/indefinite amount of time.
  87  *
  88  * Use mmput() to release the reference acquired by mmget().
  89  *
  90  * See also <Documentation/vm/active_mm.rst> for an in-depth explanation
  91  * of &mm_struct.mm_count vs &mm_struct.mm_users.
  92  */
  93 static inline void mmget(struct mm_struct *mm)
  94 {
  95         atomic_inc(&mm->mm_users);
  96 }
  97
  98 static inline bool mmget_not_zero(struct mm_struct *mm)
  99 {
 100         return atomic_inc_not_zero(&mm->mm_users);
 101 }
 102
 103 /* mmput gets rid of the mappings and all user-space */
 104 extern void mmput(struct mm_struct *);
 105 #ifdef CONFIG_MMU
 106 /* same as above but performs the slow path from the async context. Can
 107  * be called from the atomic context as well
 108  */
 109 void mmput_async(struct mm_struct *);
 110 #endif
 111
 112 /* Grab a reference to a task's mm, if it is not already going away */
 113 extern struct mm_struct *get_task_mm(struct task_struct *task);
 114 /*
 115  * Grab a reference to a task's mm, if it is not already going away
 116  * and ptrace_may_access with the mode parameter passed to it
 117  * succeeds.
 118  */
 119 extern struct mm_struct *mm_access(struct task_struct *task, unsigned int mode);
 120 /* Remove the current tasks stale references to the old mm_struct */
 121 extern void mm_release(struct task_struct *, struct mm_struct *);
 122
 123 #ifdef CONFIG_MEMCG
 124 extern void mm_update_next_owner(struct mm_struct *mm);
 125 #else
 126 static inline void mm_update_next_owner(struct mm_struct *mm)
 127 {
 128 }
 129 #endif /* CONFIG_MEMCG */
 130
 131 #ifdef CONFIG_MMU
 132 extern void arch_pick_mmap_layout(struct mm_struct *mm,
 133                                   struct rlimit *rlim_stack);
 134 extern unsigned long
 135 arch_get_unmapped_area(struct file *, unsigned long, unsigned long,
 136                        unsigned long, unsigned long);
 137 extern unsigned long
 138 arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
 139                           unsigned long len, unsigned long pgoff,
 140                           unsigned long flags);
 141 #else
 142 static inline void arch_pick_mmap_layout(struct mm_struct *mm,
 143                                          struct rlimit *rlim_stack) {}
 144 #endif
 145
 146 static inline bool in_vfork(struct task_struct *tsk)
 147 {
 148         bool ret;
 149
 150         /*
 151          * need RCU to access ->real_parent if CLONE_VM was used along with
 152          * CLONE_PARENT.
 153          *
 154          * We check real_parent->mm == tsk->mm because CLONE_VFORK does not
 155          * imply CLONE_VM
 156          *
 157          * CLONE_VFORK can be used with CLONE_PARENT/CLONE_THREAD and thus
 158          * ->real_parent is not necessarily the task doing vfork(), so in
 159          * theory we can't rely on task_lock() if we want to dereference it.
 160          *
 161          * And in this case we can't trust the real_parent->mm == tsk->mm
 162          * check, it can be false negative. But we do not care, if init or
 163          * another oom-unkillable task does this it should blame itself.
 164          */
 165         rcu_read_lock();
 166         ret = tsk->vfork_done && tsk->real_parent->mm == tsk->mm;
 167         rcu_read_unlock();
 168
 169         return ret;
 170 }
 171
 172 /*
 173  * Applies per-task gfp context to the given allocation flags.
 174  * PF_MEMALLOC_NOIO implies GFP_NOIO
 175  * PF_MEMALLOC_NOFS implies GFP_NOFS
 176  * PF_MEMALLOC_NOCMA implies no allocation from CMA region.
 177  */
 178 static inline gfp_t current_gfp_context(gfp_t flags)
 179 {
 180         if (unlikely(current->flags &
 181                      (PF_MEMALLOC_NOIO | PF_MEMALLOC_NOFS | PF_MEMALLOC_NOCMA))) {
 182                 /*
 183                  * NOIO implies both NOIO and NOFS and it is a weaker context
 184                  * so always make sure it makes precedence
 185                  */
 186                 if (current->flags & PF_MEMALLOC_NOIO)
 187                         flags &= ~(__GFP_IO | __GFP_FS);
 188                 else if (current->flags & PF_MEMALLOC_NOFS)
 189                         flags &= ~__GFP_FS;
 190 #ifdef CONFIG_CMA
 191                 if (current->flags & PF_MEMALLOC_NOCMA)
 192                         flags &= ~__GFP_MOVABLE;
 193 #endif
 194         }
 195         return flags;
 196 }
 197
 198 #ifdef CONFIG_LOCKDEP
 199 extern void __fs_reclaim_acquire(void);
 200 extern void __fs_reclaim_release(void);
 201 extern void fs_reclaim_acquire(gfp_t gfp_mask);
 202 extern void fs_reclaim_release(gfp_t gfp_mask);
 203 #else
 204 static inline void __fs_reclaim_acquire(void) { }
 205 static inline void __fs_reclaim_release(void) { }
 206 static inline void fs_reclaim_acquire(gfp_t gfp_mask) { }
 207 static inline void fs_reclaim_release(gfp_t gfp_mask) { }
 208 #endif
 209
 210 /**
 211  * memalloc_noio_save - Marks implicit GFP_NOIO allocation scope.
 212  *
 213  * This functions marks the beginning of the GFP_NOIO allocation scope.
 214  * All further allocations will implicitly drop __GFP_IO flag and so
 215  * they are safe for the IO critical section from the allocation recursion
 216  * point of view. Use memalloc_noio_restore to end the scope with flags
 217  * returned by this function.
 218  *
 219  * This function is safe to be used from any context.
 220  */
 221 static inline unsigned int memalloc_noio_save(void)
 222 {
 223         unsigned int flags = current->flags & PF_MEMALLOC_NOIO;
 224         current->flags |= PF_MEMALLOC_NOIO;
 225         return flags;
 226 }
 227
 228 /**
 229  * memalloc_noio_restore - Ends the implicit GFP_NOIO scope.
 230  * @flags: Flags to restore.
 231  *
 232  * Ends the implicit GFP_NOIO scope started by memalloc_noio_save function.
 233  * Always make sure that that the given flags is the return value from the
 234  * pairing memalloc_noio_save call.
 235  */
 236 static inline void memalloc_noio_restore(unsigned int flags)
 237 {
 238         current->flags = (current->flags & ~PF_MEMALLOC_NOIO) | flags;
 239 }
 240
 241 /**
 242  * memalloc_nofs_save - Marks implicit GFP_NOFS allocation scope.
 243  *
 244  * This functions marks the beginning of the GFP_NOFS allocation scope.
 245  * All further allocations will implicitly drop __GFP_FS flag and so
 246  * they are safe for the FS critical section from the allocation recursion
 247  * point of view. Use memalloc_nofs_restore to end the scope with flags
 248  * returned by this function.
 249  *
 250  * This function is safe to be used from any context.
 251  */
 252 static inline unsigned int memalloc_nofs_save(void)
 253 {
 254         unsigned int flags = current->flags & PF_MEMALLOC_NOFS;
 255         current->flags |= PF_MEMALLOC_NOFS;
 256         return flags;
 257 }
 258
 259 /**
 260  * memalloc_nofs_restore - Ends the implicit GFP_NOFS scope.
 261  * @flags: Flags to restore.
 262  *
 263  * Ends the implicit GFP_NOFS scope started by memalloc_nofs_save function.
 264  * Always make sure that that the given flags is the return value from the
 265  * pairing memalloc_nofs_save call.
 266  */
 267 static inline void memalloc_nofs_restore(unsigned int flags)
 268 {
 269         current->flags = (current->flags & ~PF_MEMALLOC_NOFS) | flags;
 270 }
 271
 272 static inline unsigned int memalloc_noreclaim_save(void)
 273 {
 274         unsigned int flags = current->flags & PF_MEMALLOC;
 275         current->flags |= PF_MEMALLOC;
 276         return flags;
 277 }
 278
 279 static inline void memalloc_noreclaim_restore(unsigned int flags)
 280 {
 281         current->flags = (current->flags & ~PF_MEMALLOC) | flags;
 282 }
 283
 284 #ifdef CONFIG_CMA
 285 static inline unsigned int memalloc_nocma_save(void)
 286 {
 287         unsigned int flags = current->flags & PF_MEMALLOC_NOCMA;
 288
 289         current->flags |= PF_MEMALLOC_NOCMA;
 290         return flags;
 291 }
 292
 293 static inline void memalloc_nocma_restore(unsigned int flags)
 294 {
 295         current->flags = (current->flags & ~PF_MEMALLOC_NOCMA) | flags;
 296 }
 297 #else
 298 static inline unsigned int memalloc_nocma_save(void)
 299 {
 300         return 0;
 301 }
 302
 303 static inline void memalloc_nocma_restore(unsigned int flags)
 304 {
 305 }
 306 #endif
 307
 308 #ifdef CONFIG_MEMCG
 309 /**
 310  * memalloc_use_memcg - Starts the remote memcg charging scope.
 311  * @memcg: memcg to charge.
 312  *
 313  * This function marks the beginning of the remote memcg charging scope. All the
 314  * __GFP_ACCOUNT allocations till the end of the scope will be charged to the
 315  * given memcg.
 316  *
 317  * NOTE: This function is not nesting safe.
 318  */
 319 static inline void memalloc_use_memcg(struct mem_cgroup *memcg)
 320 {
 321         WARN_ON_ONCE(current->active_memcg);
 322         current->active_memcg = memcg;
 323 }
 324
 325 /**
 326  * memalloc_unuse_memcg - Ends the remote memcg charging scope.
 327  *
 328  * This function marks the end of the remote memcg charging scope started by
 329  * memalloc_use_memcg().
 330  */
 331 static inline void memalloc_unuse_memcg(void)
 332 {
 333         current->active_memcg = NULL;
 334 }
 335 #else
 336 static inline void memalloc_use_memcg(struct mem_cgroup *memcg)
 337 {
 338 }
 339
 340 static inline void memalloc_unuse_memcg(void)
 341 {
 342 }
 343 #endif
 344
 345 #ifdef CONFIG_MEMBARRIER
 346 enum {
 347         MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY                = (1U << 0),
 348         MEMBARRIER_STATE_PRIVATE_EXPEDITED                      = (1U << 1),
 349         MEMBARRIER_STATE_GLOBAL_EXPEDITED_READY                 = (1U << 2),
 350         MEMBARRIER_STATE_GLOBAL_EXPEDITED                       = (1U << 3),
 351         MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE_READY      = (1U << 4),
 352         MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE            = (1U << 5),
 353 };
 354
 355 enum {
 356         MEMBARRIER_FLAG_SYNC_CORE       = (1U << 0),
 357 };
 358
 359 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
 360 #include <asm/membarrier.h>
 361 #endif
 362
 363 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
 364 {
 365         if (likely(!(atomic_read(&mm->membarrier_state) &
 366                      MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE)))
 367                 return;
 368         sync_core_before_usermode();
 369 }
 370
 371 static inline void membarrier_execve(struct task_struct *t)
 372 {
 373         atomic_set(&t->mm->membarrier_state, 0);
 374 }
 375 #else
 376 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
 377 static inline void membarrier_arch_switch_mm(struct mm_struct *prev,
 378                                              struct mm_struct *next,
 379                                              struct task_struct *tsk)
 380 {
 381 }
 382 #endif
 383 static inline void membarrier_execve(struct task_struct *t)
 384 {
 385 }
 386 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
 387 {
 388 }
 389 #endif
 390
 391 #endif /* _LINUX_SCHED_MM_H */