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#include <linux/sched/coredump.h>
 12
 13/*
 14 * Routines for handling mm_structs
 15 */
 16extern struct mm_struct *mm_alloc(void);
 17
 18/**
 19 * mmgrab() - Pin a &struct mm_struct.
 20 * @mm: The &struct mm_struct to pin.
 21 *
 22 * Make sure that @mm will not get freed even after the owning task
 23 * exits. This doesn't guarantee that the associated address space
 24 * will still exist later on and mmget_not_zero() has to be used before
 25 * accessing it.
 26 *
 27 * This is a preferred way to pin @mm for a longer/unbounded amount
 28 * of time.
 29 *
 30 * Use mmdrop() to release the reference acquired by mmgrab().
 31 *
 32 * See also <Documentation/mm/active_mm.rst> for an in-depth explanation
 33 * of &mm_struct.mm_count vs &mm_struct.mm_users.
 34 */
 35static inline void mmgrab(struct mm_struct *mm)
 36{
 37	atomic_inc(&mm->mm_count);
 38}
 39
 40static inline void smp_mb__after_mmgrab(void)
 41{
 42	smp_mb__after_atomic();
 43}
 44
 45extern void __mmdrop(struct mm_struct *mm);
 46
 47static inline void mmdrop(struct mm_struct *mm)
 48{
 49	/*
 50	 * The implicit full barrier implied by atomic_dec_and_test() is
 51	 * required by the membarrier system call before returning to
 52	 * user-space, after storing to rq->curr.
 53	 */
 54	if (unlikely(atomic_dec_and_test(&mm->mm_count)))
 55		__mmdrop(mm);
 56}
 57
 58#ifdef CONFIG_PREEMPT_RT
 59/*
 60 * RCU callback for delayed mm drop. Not strictly RCU, but call_rcu() is
 61 * by far the least expensive way to do that.
 62 */
 63static inline void __mmdrop_delayed(struct rcu_head *rhp)
 64{
 65	struct mm_struct *mm = container_of(rhp, struct mm_struct, delayed_drop);
 66
 67	__mmdrop(mm);
 68}
 69
 70/*
 71 * Invoked from finish_task_switch(). Delegates the heavy lifting on RT
 72 * kernels via RCU.
 73 */
 74static inline void mmdrop_sched(struct mm_struct *mm)
 75{
 76	/* Provides a full memory barrier. See mmdrop() */
 77	if (atomic_dec_and_test(&mm->mm_count))
 78		call_rcu(&mm->delayed_drop, __mmdrop_delayed);
 79}
 80#else
 81static inline void mmdrop_sched(struct mm_struct *mm)
 82{
 83	mmdrop(mm);
 84}
 85#endif
 86
 87/* Helpers for lazy TLB mm refcounting */
 88static inline void mmgrab_lazy_tlb(struct mm_struct *mm)
 89{
 90	if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_REFCOUNT))
 91		mmgrab(mm);
 92}
 93
 94static inline void mmdrop_lazy_tlb(struct mm_struct *mm)
 95{
 96	if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_REFCOUNT)) {
 97		mmdrop(mm);
 98	} else {
 99		/*
100		 * mmdrop_lazy_tlb must provide a full memory barrier, see the
101		 * membarrier comment finish_task_switch which relies on this.
102		 */
103		smp_mb();
104	}
105}
106
107static inline void mmdrop_lazy_tlb_sched(struct mm_struct *mm)
108{
109	if (IS_ENABLED(CONFIG_MMU_LAZY_TLB_REFCOUNT))
110		mmdrop_sched(mm);
111	else
112		smp_mb(); /* see mmdrop_lazy_tlb() above */
113}
114
115/**
116 * mmget() - Pin the address space associated with a &struct mm_struct.
117 * @mm: The address space to pin.
118 *
119 * Make sure that the address space of the given &struct mm_struct doesn't
120 * go away. This does not protect against parts of the address space being
121 * modified or freed, however.
122 *
123 * Never use this function to pin this address space for an
124 * unbounded/indefinite amount of time.
125 *
126 * Use mmput() to release the reference acquired by mmget().
127 *
128 * See also <Documentation/mm/active_mm.rst> for an in-depth explanation
129 * of &mm_struct.mm_count vs &mm_struct.mm_users.
130 */
131static inline void mmget(struct mm_struct *mm)
132{
133	atomic_inc(&mm->mm_users);
134}
135
136static inline bool mmget_not_zero(struct mm_struct *mm)
137{
138	return atomic_inc_not_zero(&mm->mm_users);
139}
140
141/* mmput gets rid of the mappings and all user-space */
142extern void mmput(struct mm_struct *);
143#if defined(CONFIG_MMU) || defined(CONFIG_FUTEX_PRIVATE_HASH)
144/* same as above but performs the slow path from the async context. Can
145 * be called from the atomic context as well
146 */
147void mmput_async(struct mm_struct *);
148#endif
149
150/* Grab a reference to a task's mm, if it is not already going away */
151extern struct mm_struct *get_task_mm(struct task_struct *task);
152/*
153 * Grab a reference to a task's mm, if it is not already going away
154 * and ptrace_may_access with the mode parameter passed to it
155 * succeeds.
156 */
157extern struct mm_struct *mm_access(struct task_struct *task, unsigned int mode);
158/* Remove the current tasks stale references to the old mm_struct on exit() */
159extern void exit_mm_release(struct task_struct *, struct mm_struct *);
160/* Remove the current tasks stale references to the old mm_struct on exec() */
161extern void exec_mm_release(struct task_struct *, struct mm_struct *);
162
163#ifdef CONFIG_MEMCG
164extern void mm_update_next_owner(struct mm_struct *mm);
165#else
166static inline void mm_update_next_owner(struct mm_struct *mm)
167{
168}
169#endif /* CONFIG_MEMCG */
170
171#ifdef CONFIG_MMU
172#ifndef arch_get_mmap_end
173#define arch_get_mmap_end(addr, len, flags)	(TASK_SIZE)
174#endif
175
176#ifndef arch_get_mmap_base
177#define arch_get_mmap_base(addr, base) (base)
178#endif
179
180extern void arch_pick_mmap_layout(struct mm_struct *mm,
181				  const struct rlimit *rlim_stack);
182
183unsigned long
184arch_get_unmapped_area(struct file *filp, unsigned long addr,
185		       unsigned long len, unsigned long pgoff,
186		       unsigned long flags, vm_flags_t vm_flags);
187unsigned long
188arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
189			       unsigned long len, unsigned long pgoff,
190			       unsigned long flags, vm_flags_t);
191
192unsigned long mm_get_unmapped_area(struct file *filp, unsigned long addr,
193				   unsigned long len, unsigned long pgoff,
194				   unsigned long flags);
195
196unsigned long mm_get_unmapped_area_vmflags(struct file *filp,
197					   unsigned long addr,
198					   unsigned long len,
199					   unsigned long pgoff,
200					   unsigned long flags,
201					   vm_flags_t vm_flags);
202
203unsigned long
204generic_get_unmapped_area(struct file *filp, unsigned long addr,
205			  unsigned long len, unsigned long pgoff,
206			  unsigned long flags, vm_flags_t vm_flags);
207unsigned long
208generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
209				  unsigned long len, unsigned long pgoff,
210				  unsigned long flags, vm_flags_t vm_flags);
211#else
212static inline void arch_pick_mmap_layout(struct mm_struct *mm,
213					 const struct rlimit *rlim_stack) {}
214#endif
215
216static inline bool in_vfork(struct task_struct *tsk)
217{
218	bool ret;
219
220	/*
221	 * need RCU to access ->real_parent if CLONE_VM was used along with
222	 * CLONE_PARENT.
223	 *
224	 * We check real_parent->mm == tsk->mm because CLONE_VFORK does not
225	 * imply CLONE_VM
226	 *
227	 * CLONE_VFORK can be used with CLONE_PARENT/CLONE_THREAD and thus
228	 * ->real_parent is not necessarily the task doing vfork(), so in
229	 * theory we can't rely on task_lock() if we want to dereference it.
230	 *
231	 * And in this case we can't trust the real_parent->mm == tsk->mm
232	 * check, it can be false negative. But we do not care, if init or
233	 * another oom-unkillable task does this it should blame itself.
234	 */
235	rcu_read_lock();
236	ret = tsk->vfork_done &&
237			rcu_dereference(tsk->real_parent)->mm == tsk->mm;
238	rcu_read_unlock();
239
240	return ret;
241}
242
243/*
244 * Applies per-task gfp context to the given allocation flags.
245 * PF_MEMALLOC_NOIO implies GFP_NOIO
246 * PF_MEMALLOC_NOFS implies GFP_NOFS
247 * PF_MEMALLOC_PIN  implies !GFP_MOVABLE
248 */
249static inline gfp_t current_gfp_context(gfp_t flags)
250{
251	unsigned int pflags = READ_ONCE(current->flags);
252
253	if (unlikely(pflags & (PF_MEMALLOC_NOIO | PF_MEMALLOC_NOFS | PF_MEMALLOC_PIN))) {
254		/*
255		 * NOIO implies both NOIO and NOFS and it is a weaker context
256		 * so always make sure it makes precedence
257		 */
258		if (pflags & PF_MEMALLOC_NOIO)
259			flags &= ~(__GFP_IO | __GFP_FS);
260		else if (pflags & PF_MEMALLOC_NOFS)
261			flags &= ~__GFP_FS;
262
263		if (pflags & PF_MEMALLOC_PIN)
264			flags &= ~__GFP_MOVABLE;
265	}
266	return flags;
267}
268
269#ifdef CONFIG_LOCKDEP
270extern void __fs_reclaim_acquire(unsigned long ip);
271extern void __fs_reclaim_release(unsigned long ip);
272extern void fs_reclaim_acquire(gfp_t gfp_mask);
273extern void fs_reclaim_release(gfp_t gfp_mask);
274#else
275static inline void __fs_reclaim_acquire(unsigned long ip) { }
276static inline void __fs_reclaim_release(unsigned long ip) { }
277static inline void fs_reclaim_acquire(gfp_t gfp_mask) { }
278static inline void fs_reclaim_release(gfp_t gfp_mask) { }
279#endif
280
281/* Any memory-allocation retry loop should use
282 * memalloc_retry_wait(), and pass the flags for the most
283 * constrained allocation attempt that might have failed.
284 * This provides useful documentation of where loops are,
285 * and a central place to fine tune the waiting as the MM
286 * implementation changes.
287 */
288static inline void memalloc_retry_wait(gfp_t gfp_flags)
289{
290	/* We use io_schedule_timeout because waiting for memory
291	 * typically included waiting for dirty pages to be
292	 * written out, which requires IO.
293	 */
294	__set_current_state(TASK_UNINTERRUPTIBLE);
295	gfp_flags = current_gfp_context(gfp_flags);
296	if (gfpflags_allow_blocking(gfp_flags) &&
297	    !(gfp_flags & __GFP_NORETRY))
298		/* Probably waited already, no need for much more */
299		io_schedule_timeout(1);
300	else
301		/* Probably didn't wait, and has now released a lock,
302		 * so now is a good time to wait
303		 */
304		io_schedule_timeout(HZ/50);
305}
306
307/**
308 * might_alloc - Mark possible allocation sites
309 * @gfp_mask: gfp_t flags that would be used to allocate
310 *
311 * Similar to might_sleep() and other annotations, this can be used in functions
312 * that might allocate, but often don't. Compiles to nothing without
313 * CONFIG_LOCKDEP. Includes a conditional might_sleep() if @gfp allows blocking.
314 */
315static inline void might_alloc(gfp_t gfp_mask)
316{
317	fs_reclaim_acquire(gfp_mask);
318	fs_reclaim_release(gfp_mask);
319
320	if (current->flags & PF_MEMALLOC)
321		return;
322
323	might_sleep_if(gfpflags_allow_blocking(gfp_mask));
324}
325
326/**
327 * memalloc_flags_save - Add a PF_* flag to current->flags, save old value
328 * @flags: Flags to add.
329 *
330 * This allows PF_* flags to be conveniently added, irrespective of current
331 * value, and then the old version restored with memalloc_flags_restore().
332 */
333static inline unsigned memalloc_flags_save(unsigned flags)
334{
335	unsigned oldflags = ~current->flags & flags;
336	current->flags |= flags;
337	return oldflags;
338}
339
340static inline void memalloc_flags_restore(unsigned flags)
341{
342	current->flags &= ~flags;
343}
344
345/**
346 * memalloc_noio_save - Marks implicit GFP_NOIO allocation scope.
347 *
348 * This functions marks the beginning of the GFP_NOIO allocation scope.
349 * All further allocations will implicitly drop __GFP_IO flag and so
350 * they are safe for the IO critical section from the allocation recursion
351 * point of view. Use memalloc_noio_restore to end the scope with flags
352 * returned by this function.
353 *
354 * Context: This function is safe to be used from any context.
355 * Return: The saved flags to be passed to memalloc_noio_restore.
356 */
357static inline unsigned int memalloc_noio_save(void)
358{
359	return memalloc_flags_save(PF_MEMALLOC_NOIO);
360}
361
362/**
363 * memalloc_noio_restore - Ends the implicit GFP_NOIO scope.
364 * @flags: Flags to restore.
365 *
366 * Ends the implicit GFP_NOIO scope started by memalloc_noio_save function.
367 * Always make sure that the given flags is the return value from the
368 * pairing memalloc_noio_save call.
369 */
370static inline void memalloc_noio_restore(unsigned int flags)
371{
372	memalloc_flags_restore(flags);
373}
374
375/**
376 * memalloc_nofs_save - Marks implicit GFP_NOFS allocation scope.
377 *
378 * This functions marks the beginning of the GFP_NOFS allocation scope.
379 * All further allocations will implicitly drop __GFP_FS flag and so
380 * they are safe for the FS critical section from the allocation recursion
381 * point of view. Use memalloc_nofs_restore to end the scope with flags
382 * returned by this function.
383 *
384 * Context: This function is safe to be used from any context.
385 * Return: The saved flags to be passed to memalloc_nofs_restore.
386 */
387static inline unsigned int memalloc_nofs_save(void)
388{
389	return memalloc_flags_save(PF_MEMALLOC_NOFS);
390}
391
392/**
393 * memalloc_nofs_restore - Ends the implicit GFP_NOFS scope.
394 * @flags: Flags to restore.
395 *
396 * Ends the implicit GFP_NOFS scope started by memalloc_nofs_save function.
397 * Always make sure that the given flags is the return value from the
398 * pairing memalloc_nofs_save call.
399 */
400static inline void memalloc_nofs_restore(unsigned int flags)
401{
402	memalloc_flags_restore(flags);
403}
404
405/**
406 * memalloc_noreclaim_save - Marks implicit __GFP_MEMALLOC scope.
407 *
408 * This function marks the beginning of the __GFP_MEMALLOC allocation scope.
409 * All further allocations will implicitly add the __GFP_MEMALLOC flag, which
410 * prevents entering reclaim and allows access to all memory reserves. This
411 * should only be used when the caller guarantees the allocation will allow more
412 * memory to be freed very shortly, i.e. it needs to allocate some memory in
413 * the process of freeing memory, and cannot reclaim due to potential recursion.
414 *
415 * Users of this scope have to be extremely careful to not deplete the reserves
416 * completely and implement a throttling mechanism which controls the
417 * consumption of the reserve based on the amount of freed memory. Usage of a
418 * pre-allocated pool (e.g. mempool) should be always considered before using
419 * this scope.
420 *
421 * Individual allocations under the scope can opt out using __GFP_NOMEMALLOC
422 *
423 * Context: This function should not be used in an interrupt context as that one
424 *          does not give PF_MEMALLOC access to reserves.
425 *          See __gfp_pfmemalloc_flags().
426 * Return: The saved flags to be passed to memalloc_noreclaim_restore.
427 */
428static inline unsigned int memalloc_noreclaim_save(void)
429{
430	return memalloc_flags_save(PF_MEMALLOC);
431}
432
433/**
434 * memalloc_noreclaim_restore - Ends the implicit __GFP_MEMALLOC scope.
435 * @flags: Flags to restore.
436 *
437 * Ends the implicit __GFP_MEMALLOC scope started by memalloc_noreclaim_save
438 * function. Always make sure that the given flags is the return value from the
439 * pairing memalloc_noreclaim_save call.
440 */
441static inline void memalloc_noreclaim_restore(unsigned int flags)
442{
443	memalloc_flags_restore(flags);
444}
445
446/**
447 * memalloc_pin_save - Marks implicit ~__GFP_MOVABLE scope.
448 *
449 * This function marks the beginning of the ~__GFP_MOVABLE allocation scope.
450 * All further allocations will implicitly remove the __GFP_MOVABLE flag, which
451 * will constraint the allocations to zones that allow long term pinning, i.e.
452 * not ZONE_MOVABLE zones.
453 *
454 * Return: The saved flags to be passed to memalloc_pin_restore.
455 */
456static inline unsigned int memalloc_pin_save(void)
457{
458	return memalloc_flags_save(PF_MEMALLOC_PIN);
459}
460
461/**
462 * memalloc_pin_restore - Ends the implicit ~__GFP_MOVABLE scope.
463 * @flags: Flags to restore.
464 *
465 * Ends the implicit ~__GFP_MOVABLE scope started by memalloc_pin_save function.
466 * Always make sure that the given flags is the return value from the pairing
467 * memalloc_pin_save call.
468 */
469static inline void memalloc_pin_restore(unsigned int flags)
470{
471	memalloc_flags_restore(flags);
472}
473
474#ifdef CONFIG_MEMCG
475DECLARE_PER_CPU(struct mem_cgroup *, int_active_memcg);
476/**
477 * set_active_memcg - Starts the remote memcg charging scope.
478 * @memcg: memcg to charge.
479 *
480 * This function marks the beginning of the remote memcg charging scope. All the
481 * __GFP_ACCOUNT allocations till the end of the scope will be charged to the
482 * given memcg.
483 *
484 * Please, make sure that caller has a reference to the passed memcg structure,
485 * so its lifetime is guaranteed to exceed the scope between two
486 * set_active_memcg() calls.
487 *
488 * NOTE: This function can nest. Users must save the return value and
489 * reset the previous value after their own charging scope is over.
490 */
491static inline struct mem_cgroup *
492set_active_memcg(struct mem_cgroup *memcg)
493{
494	struct mem_cgroup *old;
495
496	if (!in_task()) {
497		old = this_cpu_read(int_active_memcg);
498		this_cpu_write(int_active_memcg, memcg);
499	} else {
500		old = current->active_memcg;
501		current->active_memcg = memcg;
502	}
503
504	return old;
505}
506#else
507static inline struct mem_cgroup *
508set_active_memcg(struct mem_cgroup *memcg)
509{
510	return NULL;
511}
512#endif
513
514#ifdef CONFIG_MEMBARRIER
515enum {
516	MEMBARRIER_STATE_PRIVATE_EXPEDITED_READY		= (1U << 0),
517	MEMBARRIER_STATE_PRIVATE_EXPEDITED			= (1U << 1),
518	MEMBARRIER_STATE_GLOBAL_EXPEDITED_READY			= (1U << 2),
519	MEMBARRIER_STATE_GLOBAL_EXPEDITED			= (1U << 3),
520	MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE_READY	= (1U << 4),
521	MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE		= (1U << 5),
522	MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ_READY		= (1U << 6),
523	MEMBARRIER_STATE_PRIVATE_EXPEDITED_RSEQ			= (1U << 7),
524};
525
526enum {
527	MEMBARRIER_FLAG_SYNC_CORE	= (1U << 0),
528	MEMBARRIER_FLAG_RSEQ		= (1U << 1),
529};
530
531#ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
532#include <asm/membarrier.h>
533#endif
534
535static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
536{
537	/*
538	 * The atomic_read() below prevents CSE. The following should
539	 * help the compiler generate more efficient code on architectures
540	 * where sync_core_before_usermode() is a no-op.
541	 */
542	if (!IS_ENABLED(CONFIG_ARCH_HAS_SYNC_CORE_BEFORE_USERMODE))
543		return;
544	if (current->mm != mm)
545		return;
546	if (likely(!(atomic_read(&mm->membarrier_state) &
547		     MEMBARRIER_STATE_PRIVATE_EXPEDITED_SYNC_CORE)))
548		return;
549	sync_core_before_usermode();
550}
551
552extern void membarrier_exec_mmap(struct mm_struct *mm);
553
554extern void membarrier_update_current_mm(struct mm_struct *next_mm);
555
556#else
557#ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
558static inline void membarrier_arch_switch_mm(struct mm_struct *prev,
559					     struct mm_struct *next,
560					     struct task_struct *tsk)
561{
562}
563#endif
564static inline void membarrier_exec_mmap(struct mm_struct *mm)
565{
566}
567static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
568{
569}
570static inline void membarrier_update_current_mm(struct mm_struct *next_mm)
571{
572}
573#endif
574
575#endif /* _LINUX_SCHED_MM_H */