xref: /linux/mm/mlock.c (revision ab52c59103002b49f2455371e4b9c56ba3ef1781)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *	linux/mm/mlock.c
4  *
5  *  (C) Copyright 1995 Linus Torvalds
6  *  (C) Copyright 2002 Christoph Hellwig
7  */
8 
9 #include <linux/capability.h>
10 #include <linux/mman.h>
11 #include <linux/mm.h>
12 #include <linux/sched/user.h>
13 #include <linux/swap.h>
14 #include <linux/swapops.h>
15 #include <linux/pagemap.h>
16 #include <linux/pagevec.h>
17 #include <linux/pagewalk.h>
18 #include <linux/mempolicy.h>
19 #include <linux/syscalls.h>
20 #include <linux/sched.h>
21 #include <linux/export.h>
22 #include <linux/rmap.h>
23 #include <linux/mmzone.h>
24 #include <linux/hugetlb.h>
25 #include <linux/memcontrol.h>
26 #include <linux/mm_inline.h>
27 #include <linux/secretmem.h>
28 
29 #include "internal.h"
30 
31 struct mlock_fbatch {
32 	local_lock_t lock;
33 	struct folio_batch fbatch;
34 };
35 
36 static DEFINE_PER_CPU(struct mlock_fbatch, mlock_fbatch) = {
37 	.lock = INIT_LOCAL_LOCK(lock),
38 };
39 
40 bool can_do_mlock(void)
41 {
42 	if (rlimit(RLIMIT_MEMLOCK) != 0)
43 		return true;
44 	if (capable(CAP_IPC_LOCK))
45 		return true;
46 	return false;
47 }
48 EXPORT_SYMBOL(can_do_mlock);
49 
50 /*
51  * Mlocked folios are marked with the PG_mlocked flag for efficient testing
52  * in vmscan and, possibly, the fault path; and to support semi-accurate
53  * statistics.
54  *
55  * An mlocked folio [folio_test_mlocked(folio)] is unevictable.  As such, it
56  * will be ostensibly placed on the LRU "unevictable" list (actually no such
57  * list exists), rather than the [in]active lists. PG_unevictable is set to
58  * indicate the unevictable state.
59  */
60 
61 static struct lruvec *__mlock_folio(struct folio *folio, struct lruvec *lruvec)
62 {
63 	/* There is nothing more we can do while it's off LRU */
64 	if (!folio_test_clear_lru(folio))
65 		return lruvec;
66 
67 	lruvec = folio_lruvec_relock_irq(folio, lruvec);
68 
69 	if (unlikely(folio_evictable(folio))) {
70 		/*
71 		 * This is a little surprising, but quite possible: PG_mlocked
72 		 * must have got cleared already by another CPU.  Could this
73 		 * folio be unevictable?  I'm not sure, but move it now if so.
74 		 */
75 		if (folio_test_unevictable(folio)) {
76 			lruvec_del_folio(lruvec, folio);
77 			folio_clear_unevictable(folio);
78 			lruvec_add_folio(lruvec, folio);
79 
80 			__count_vm_events(UNEVICTABLE_PGRESCUED,
81 					  folio_nr_pages(folio));
82 		}
83 		goto out;
84 	}
85 
86 	if (folio_test_unevictable(folio)) {
87 		if (folio_test_mlocked(folio))
88 			folio->mlock_count++;
89 		goto out;
90 	}
91 
92 	lruvec_del_folio(lruvec, folio);
93 	folio_clear_active(folio);
94 	folio_set_unevictable(folio);
95 	folio->mlock_count = !!folio_test_mlocked(folio);
96 	lruvec_add_folio(lruvec, folio);
97 	__count_vm_events(UNEVICTABLE_PGCULLED, folio_nr_pages(folio));
98 out:
99 	folio_set_lru(folio);
100 	return lruvec;
101 }
102 
103 static struct lruvec *__mlock_new_folio(struct folio *folio, struct lruvec *lruvec)
104 {
105 	VM_BUG_ON_FOLIO(folio_test_lru(folio), folio);
106 
107 	lruvec = folio_lruvec_relock_irq(folio, lruvec);
108 
109 	/* As above, this is a little surprising, but possible */
110 	if (unlikely(folio_evictable(folio)))
111 		goto out;
112 
113 	folio_set_unevictable(folio);
114 	folio->mlock_count = !!folio_test_mlocked(folio);
115 	__count_vm_events(UNEVICTABLE_PGCULLED, folio_nr_pages(folio));
116 out:
117 	lruvec_add_folio(lruvec, folio);
118 	folio_set_lru(folio);
119 	return lruvec;
120 }
121 
122 static struct lruvec *__munlock_folio(struct folio *folio, struct lruvec *lruvec)
123 {
124 	int nr_pages = folio_nr_pages(folio);
125 	bool isolated = false;
126 
127 	if (!folio_test_clear_lru(folio))
128 		goto munlock;
129 
130 	isolated = true;
131 	lruvec = folio_lruvec_relock_irq(folio, lruvec);
132 
133 	if (folio_test_unevictable(folio)) {
134 		/* Then mlock_count is maintained, but might undercount */
135 		if (folio->mlock_count)
136 			folio->mlock_count--;
137 		if (folio->mlock_count)
138 			goto out;
139 	}
140 	/* else assume that was the last mlock: reclaim will fix it if not */
141 
142 munlock:
143 	if (folio_test_clear_mlocked(folio)) {
144 		__zone_stat_mod_folio(folio, NR_MLOCK, -nr_pages);
145 		if (isolated || !folio_test_unevictable(folio))
146 			__count_vm_events(UNEVICTABLE_PGMUNLOCKED, nr_pages);
147 		else
148 			__count_vm_events(UNEVICTABLE_PGSTRANDED, nr_pages);
149 	}
150 
151 	/* folio_evictable() has to be checked *after* clearing Mlocked */
152 	if (isolated && folio_test_unevictable(folio) && folio_evictable(folio)) {
153 		lruvec_del_folio(lruvec, folio);
154 		folio_clear_unevictable(folio);
155 		lruvec_add_folio(lruvec, folio);
156 		__count_vm_events(UNEVICTABLE_PGRESCUED, nr_pages);
157 	}
158 out:
159 	if (isolated)
160 		folio_set_lru(folio);
161 	return lruvec;
162 }
163 
164 /*
165  * Flags held in the low bits of a struct folio pointer on the mlock_fbatch.
166  */
167 #define LRU_FOLIO 0x1
168 #define NEW_FOLIO 0x2
169 static inline struct folio *mlock_lru(struct folio *folio)
170 {
171 	return (struct folio *)((unsigned long)folio + LRU_FOLIO);
172 }
173 
174 static inline struct folio *mlock_new(struct folio *folio)
175 {
176 	return (struct folio *)((unsigned long)folio + NEW_FOLIO);
177 }
178 
179 /*
180  * mlock_folio_batch() is derived from folio_batch_move_lru(): perhaps that can
181  * make use of such folio pointer flags in future, but for now just keep it for
182  * mlock.  We could use three separate folio batches instead, but one feels
183  * better (munlocking a full folio batch does not need to drain mlocking folio
184  * batches first).
185  */
186 static void mlock_folio_batch(struct folio_batch *fbatch)
187 {
188 	struct lruvec *lruvec = NULL;
189 	unsigned long mlock;
190 	struct folio *folio;
191 	int i;
192 
193 	for (i = 0; i < folio_batch_count(fbatch); i++) {
194 		folio = fbatch->folios[i];
195 		mlock = (unsigned long)folio & (LRU_FOLIO | NEW_FOLIO);
196 		folio = (struct folio *)((unsigned long)folio - mlock);
197 		fbatch->folios[i] = folio;
198 
199 		if (mlock & LRU_FOLIO)
200 			lruvec = __mlock_folio(folio, lruvec);
201 		else if (mlock & NEW_FOLIO)
202 			lruvec = __mlock_new_folio(folio, lruvec);
203 		else
204 			lruvec = __munlock_folio(folio, lruvec);
205 	}
206 
207 	if (lruvec)
208 		unlock_page_lruvec_irq(lruvec);
209 	folios_put(fbatch);
210 }
211 
212 void mlock_drain_local(void)
213 {
214 	struct folio_batch *fbatch;
215 
216 	local_lock(&mlock_fbatch.lock);
217 	fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
218 	if (folio_batch_count(fbatch))
219 		mlock_folio_batch(fbatch);
220 	local_unlock(&mlock_fbatch.lock);
221 }
222 
223 void mlock_drain_remote(int cpu)
224 {
225 	struct folio_batch *fbatch;
226 
227 	WARN_ON_ONCE(cpu_online(cpu));
228 	fbatch = &per_cpu(mlock_fbatch.fbatch, cpu);
229 	if (folio_batch_count(fbatch))
230 		mlock_folio_batch(fbatch);
231 }
232 
233 bool need_mlock_drain(int cpu)
234 {
235 	return folio_batch_count(&per_cpu(mlock_fbatch.fbatch, cpu));
236 }
237 
238 /**
239  * mlock_folio - mlock a folio already on (or temporarily off) LRU
240  * @folio: folio to be mlocked.
241  */
242 void mlock_folio(struct folio *folio)
243 {
244 	struct folio_batch *fbatch;
245 
246 	local_lock(&mlock_fbatch.lock);
247 	fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
248 
249 	if (!folio_test_set_mlocked(folio)) {
250 		int nr_pages = folio_nr_pages(folio);
251 
252 		zone_stat_mod_folio(folio, NR_MLOCK, nr_pages);
253 		__count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
254 	}
255 
256 	folio_get(folio);
257 	if (!folio_batch_add(fbatch, mlock_lru(folio)) ||
258 	    folio_test_large(folio) || lru_cache_disabled())
259 		mlock_folio_batch(fbatch);
260 	local_unlock(&mlock_fbatch.lock);
261 }
262 
263 /**
264  * mlock_new_folio - mlock a newly allocated folio not yet on LRU
265  * @folio: folio to be mlocked, either normal or a THP head.
266  */
267 void mlock_new_folio(struct folio *folio)
268 {
269 	struct folio_batch *fbatch;
270 	int nr_pages = folio_nr_pages(folio);
271 
272 	local_lock(&mlock_fbatch.lock);
273 	fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
274 	folio_set_mlocked(folio);
275 
276 	zone_stat_mod_folio(folio, NR_MLOCK, nr_pages);
277 	__count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
278 
279 	folio_get(folio);
280 	if (!folio_batch_add(fbatch, mlock_new(folio)) ||
281 	    folio_test_large(folio) || lru_cache_disabled())
282 		mlock_folio_batch(fbatch);
283 	local_unlock(&mlock_fbatch.lock);
284 }
285 
286 /**
287  * munlock_folio - munlock a folio
288  * @folio: folio to be munlocked, either normal or a THP head.
289  */
290 void munlock_folio(struct folio *folio)
291 {
292 	struct folio_batch *fbatch;
293 
294 	local_lock(&mlock_fbatch.lock);
295 	fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
296 	/*
297 	 * folio_test_clear_mlocked(folio) must be left to __munlock_folio(),
298 	 * which will check whether the folio is multiply mlocked.
299 	 */
300 	folio_get(folio);
301 	if (!folio_batch_add(fbatch, folio) ||
302 	    folio_test_large(folio) || lru_cache_disabled())
303 		mlock_folio_batch(fbatch);
304 	local_unlock(&mlock_fbatch.lock);
305 }
306 
307 static inline unsigned int folio_mlock_step(struct folio *folio,
308 		pte_t *pte, unsigned long addr, unsigned long end)
309 {
310 	unsigned int count, i, nr = folio_nr_pages(folio);
311 	unsigned long pfn = folio_pfn(folio);
312 	pte_t ptent = ptep_get(pte);
313 
314 	if (!folio_test_large(folio))
315 		return 1;
316 
317 	count = pfn + nr - pte_pfn(ptent);
318 	count = min_t(unsigned int, count, (end - addr) >> PAGE_SHIFT);
319 
320 	for (i = 0; i < count; i++, pte++) {
321 		pte_t entry = ptep_get(pte);
322 
323 		if (!pte_present(entry))
324 			break;
325 		if (pte_pfn(entry) - pfn >= nr)
326 			break;
327 	}
328 
329 	return i;
330 }
331 
332 static inline bool allow_mlock_munlock(struct folio *folio,
333 		struct vm_area_struct *vma, unsigned long start,
334 		unsigned long end, unsigned int step)
335 {
336 	/*
337 	 * For unlock, allow munlock large folio which is partially
338 	 * mapped to VMA. As it's possible that large folio is
339 	 * mlocked and VMA is split later.
340 	 *
341 	 * During memory pressure, such kind of large folio can
342 	 * be split. And the pages are not in VM_LOCKed VMA
343 	 * can be reclaimed.
344 	 */
345 	if (!(vma->vm_flags & VM_LOCKED))
346 		return true;
347 
348 	/* folio_within_range() cannot take KSM, but any small folio is OK */
349 	if (!folio_test_large(folio))
350 		return true;
351 
352 	/* folio not in range [start, end), skip mlock */
353 	if (!folio_within_range(folio, vma, start, end))
354 		return false;
355 
356 	/* folio is not fully mapped, skip mlock */
357 	if (step != folio_nr_pages(folio))
358 		return false;
359 
360 	return true;
361 }
362 
363 static int mlock_pte_range(pmd_t *pmd, unsigned long addr,
364 			   unsigned long end, struct mm_walk *walk)
365 
366 {
367 	struct vm_area_struct *vma = walk->vma;
368 	spinlock_t *ptl;
369 	pte_t *start_pte, *pte;
370 	pte_t ptent;
371 	struct folio *folio;
372 	unsigned int step = 1;
373 	unsigned long start = addr;
374 
375 	ptl = pmd_trans_huge_lock(pmd, vma);
376 	if (ptl) {
377 		if (!pmd_present(*pmd))
378 			goto out;
379 		if (is_huge_zero_pmd(*pmd))
380 			goto out;
381 		folio = pmd_folio(*pmd);
382 		if (vma->vm_flags & VM_LOCKED)
383 			mlock_folio(folio);
384 		else
385 			munlock_folio(folio);
386 		goto out;
387 	}
388 
389 	start_pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
390 	if (!start_pte) {
391 		walk->action = ACTION_AGAIN;
392 		return 0;
393 	}
394 
395 	for (pte = start_pte; addr != end; pte++, addr += PAGE_SIZE) {
396 		ptent = ptep_get(pte);
397 		if (!pte_present(ptent))
398 			continue;
399 		folio = vm_normal_folio(vma, addr, ptent);
400 		if (!folio || folio_is_zone_device(folio))
401 			continue;
402 
403 		step = folio_mlock_step(folio, pte, addr, end);
404 		if (!allow_mlock_munlock(folio, vma, start, end, step))
405 			goto next_entry;
406 
407 		if (vma->vm_flags & VM_LOCKED)
408 			mlock_folio(folio);
409 		else
410 			munlock_folio(folio);
411 
412 next_entry:
413 		pte += step - 1;
414 		addr += (step - 1) << PAGE_SHIFT;
415 	}
416 	pte_unmap(start_pte);
417 out:
418 	spin_unlock(ptl);
419 	cond_resched();
420 	return 0;
421 }
422 
423 /*
424  * mlock_vma_pages_range() - mlock any pages already in the range,
425  *                           or munlock all pages in the range.
426  * @vma - vma containing range to be mlock()ed or munlock()ed
427  * @start - start address in @vma of the range
428  * @end - end of range in @vma
429  * @newflags - the new set of flags for @vma.
430  *
431  * Called for mlock(), mlock2() and mlockall(), to set @vma VM_LOCKED;
432  * called for munlock() and munlockall(), to clear VM_LOCKED from @vma.
433  */
434 static void mlock_vma_pages_range(struct vm_area_struct *vma,
435 	unsigned long start, unsigned long end, vm_flags_t newflags)
436 {
437 	static const struct mm_walk_ops mlock_walk_ops = {
438 		.pmd_entry = mlock_pte_range,
439 		.walk_lock = PGWALK_WRLOCK_VERIFY,
440 	};
441 
442 	/*
443 	 * There is a slight chance that concurrent page migration,
444 	 * or page reclaim finding a page of this now-VM_LOCKED vma,
445 	 * will call mlock_vma_folio() and raise page's mlock_count:
446 	 * double counting, leaving the page unevictable indefinitely.
447 	 * Communicate this danger to mlock_vma_folio() with VM_IO,
448 	 * which is a VM_SPECIAL flag not allowed on VM_LOCKED vmas.
449 	 * mmap_lock is held in write mode here, so this weird
450 	 * combination should not be visible to other mmap_lock users;
451 	 * but WRITE_ONCE so rmap walkers must see VM_IO if VM_LOCKED.
452 	 */
453 	if (newflags & VM_LOCKED)
454 		newflags |= VM_IO;
455 	vma_start_write(vma);
456 	vm_flags_reset_once(vma, newflags);
457 
458 	lru_add_drain();
459 	walk_page_range(vma->vm_mm, start, end, &mlock_walk_ops, NULL);
460 	lru_add_drain();
461 
462 	if (newflags & VM_IO) {
463 		newflags &= ~VM_IO;
464 		vm_flags_reset_once(vma, newflags);
465 	}
466 }
467 
468 /*
469  * mlock_fixup  - handle mlock[all]/munlock[all] requests.
470  *
471  * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
472  * munlock is a no-op.  However, for some special vmas, we go ahead and
473  * populate the ptes.
474  *
475  * For vmas that pass the filters, merge/split as appropriate.
476  */
477 static int mlock_fixup(struct vma_iterator *vmi, struct vm_area_struct *vma,
478 	       struct vm_area_struct **prev, unsigned long start,
479 	       unsigned long end, vm_flags_t newflags)
480 {
481 	struct mm_struct *mm = vma->vm_mm;
482 	int nr_pages;
483 	int ret = 0;
484 	vm_flags_t oldflags = vma->vm_flags;
485 
486 	if (newflags == oldflags || (oldflags & VM_SPECIAL) ||
487 	    is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm) ||
488 	    vma_is_dax(vma) || vma_is_secretmem(vma))
489 		/* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
490 		goto out;
491 
492 	vma = vma_modify_flags(vmi, *prev, vma, start, end, newflags);
493 	if (IS_ERR(vma)) {
494 		ret = PTR_ERR(vma);
495 		goto out;
496 	}
497 
498 	/*
499 	 * Keep track of amount of locked VM.
500 	 */
501 	nr_pages = (end - start) >> PAGE_SHIFT;
502 	if (!(newflags & VM_LOCKED))
503 		nr_pages = -nr_pages;
504 	else if (oldflags & VM_LOCKED)
505 		nr_pages = 0;
506 	mm->locked_vm += nr_pages;
507 
508 	/*
509 	 * vm_flags is protected by the mmap_lock held in write mode.
510 	 * It's okay if try_to_unmap_one unmaps a page just after we
511 	 * set VM_LOCKED, populate_vma_page_range will bring it back.
512 	 */
513 	if ((newflags & VM_LOCKED) && (oldflags & VM_LOCKED)) {
514 		/* No work to do, and mlocking twice would be wrong */
515 		vma_start_write(vma);
516 		vm_flags_reset(vma, newflags);
517 	} else {
518 		mlock_vma_pages_range(vma, start, end, newflags);
519 	}
520 out:
521 	*prev = vma;
522 	return ret;
523 }
524 
525 static int apply_vma_lock_flags(unsigned long start, size_t len,
526 				vm_flags_t flags)
527 {
528 	unsigned long nstart, end, tmp;
529 	struct vm_area_struct *vma, *prev;
530 	VMA_ITERATOR(vmi, current->mm, start);
531 
532 	VM_BUG_ON(offset_in_page(start));
533 	VM_BUG_ON(len != PAGE_ALIGN(len));
534 	end = start + len;
535 	if (end < start)
536 		return -EINVAL;
537 	if (end == start)
538 		return 0;
539 	vma = vma_iter_load(&vmi);
540 	if (!vma)
541 		return -ENOMEM;
542 
543 	prev = vma_prev(&vmi);
544 	if (start > vma->vm_start)
545 		prev = vma;
546 
547 	nstart = start;
548 	tmp = vma->vm_start;
549 	for_each_vma_range(vmi, vma, end) {
550 		int error;
551 		vm_flags_t newflags;
552 
553 		if (vma->vm_start != tmp)
554 			return -ENOMEM;
555 
556 		newflags = vma->vm_flags & ~VM_LOCKED_MASK;
557 		newflags |= flags;
558 		/* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
559 		tmp = vma->vm_end;
560 		if (tmp > end)
561 			tmp = end;
562 		error = mlock_fixup(&vmi, vma, &prev, nstart, tmp, newflags);
563 		if (error)
564 			return error;
565 		tmp = vma_iter_end(&vmi);
566 		nstart = tmp;
567 	}
568 
569 	if (tmp < end)
570 		return -ENOMEM;
571 
572 	return 0;
573 }
574 
575 /*
576  * Go through vma areas and sum size of mlocked
577  * vma pages, as return value.
578  * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT)
579  * is also counted.
580  * Return value: previously mlocked page counts
581  */
582 static unsigned long count_mm_mlocked_page_nr(struct mm_struct *mm,
583 		unsigned long start, size_t len)
584 {
585 	struct vm_area_struct *vma;
586 	unsigned long count = 0;
587 	unsigned long end;
588 	VMA_ITERATOR(vmi, mm, start);
589 
590 	/* Don't overflow past ULONG_MAX */
591 	if (unlikely(ULONG_MAX - len < start))
592 		end = ULONG_MAX;
593 	else
594 		end = start + len;
595 
596 	for_each_vma_range(vmi, vma, end) {
597 		if (vma->vm_flags & VM_LOCKED) {
598 			if (start > vma->vm_start)
599 				count -= (start - vma->vm_start);
600 			if (end < vma->vm_end) {
601 				count += end - vma->vm_start;
602 				break;
603 			}
604 			count += vma->vm_end - vma->vm_start;
605 		}
606 	}
607 
608 	return count >> PAGE_SHIFT;
609 }
610 
611 /*
612  * convert get_user_pages() return value to posix mlock() error
613  */
614 static int __mlock_posix_error_return(long retval)
615 {
616 	if (retval == -EFAULT)
617 		retval = -ENOMEM;
618 	else if (retval == -ENOMEM)
619 		retval = -EAGAIN;
620 	return retval;
621 }
622 
623 static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
624 {
625 	unsigned long locked;
626 	unsigned long lock_limit;
627 	int error = -ENOMEM;
628 
629 	start = untagged_addr(start);
630 
631 	if (!can_do_mlock())
632 		return -EPERM;
633 
634 	len = PAGE_ALIGN(len + (offset_in_page(start)));
635 	start &= PAGE_MASK;
636 
637 	lock_limit = rlimit(RLIMIT_MEMLOCK);
638 	lock_limit >>= PAGE_SHIFT;
639 	locked = len >> PAGE_SHIFT;
640 
641 	if (mmap_write_lock_killable(current->mm))
642 		return -EINTR;
643 
644 	locked += current->mm->locked_vm;
645 	if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) {
646 		/*
647 		 * It is possible that the regions requested intersect with
648 		 * previously mlocked areas, that part area in "mm->locked_vm"
649 		 * should not be counted to new mlock increment count. So check
650 		 * and adjust locked count if necessary.
651 		 */
652 		locked -= count_mm_mlocked_page_nr(current->mm,
653 				start, len);
654 	}
655 
656 	/* check against resource limits */
657 	if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
658 		error = apply_vma_lock_flags(start, len, flags);
659 
660 	mmap_write_unlock(current->mm);
661 	if (error)
662 		return error;
663 
664 	error = __mm_populate(start, len, 0);
665 	if (error)
666 		return __mlock_posix_error_return(error);
667 	return 0;
668 }
669 
670 SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
671 {
672 	return do_mlock(start, len, VM_LOCKED);
673 }
674 
675 SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
676 {
677 	vm_flags_t vm_flags = VM_LOCKED;
678 
679 	if (flags & ~MLOCK_ONFAULT)
680 		return -EINVAL;
681 
682 	if (flags & MLOCK_ONFAULT)
683 		vm_flags |= VM_LOCKONFAULT;
684 
685 	return do_mlock(start, len, vm_flags);
686 }
687 
688 SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
689 {
690 	int ret;
691 
692 	start = untagged_addr(start);
693 
694 	len = PAGE_ALIGN(len + (offset_in_page(start)));
695 	start &= PAGE_MASK;
696 
697 	if (mmap_write_lock_killable(current->mm))
698 		return -EINTR;
699 	ret = apply_vma_lock_flags(start, len, 0);
700 	mmap_write_unlock(current->mm);
701 
702 	return ret;
703 }
704 
705 /*
706  * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
707  * and translate into the appropriate modifications to mm->def_flags and/or the
708  * flags for all current VMAs.
709  *
710  * There are a couple of subtleties with this.  If mlockall() is called multiple
711  * times with different flags, the values do not necessarily stack.  If mlockall
712  * is called once including the MCL_FUTURE flag and then a second time without
713  * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
714  */
715 static int apply_mlockall_flags(int flags)
716 {
717 	VMA_ITERATOR(vmi, current->mm, 0);
718 	struct vm_area_struct *vma, *prev = NULL;
719 	vm_flags_t to_add = 0;
720 
721 	current->mm->def_flags &= ~VM_LOCKED_MASK;
722 	if (flags & MCL_FUTURE) {
723 		current->mm->def_flags |= VM_LOCKED;
724 
725 		if (flags & MCL_ONFAULT)
726 			current->mm->def_flags |= VM_LOCKONFAULT;
727 
728 		if (!(flags & MCL_CURRENT))
729 			goto out;
730 	}
731 
732 	if (flags & MCL_CURRENT) {
733 		to_add |= VM_LOCKED;
734 		if (flags & MCL_ONFAULT)
735 			to_add |= VM_LOCKONFAULT;
736 	}
737 
738 	for_each_vma(vmi, vma) {
739 		vm_flags_t newflags;
740 
741 		newflags = vma->vm_flags & ~VM_LOCKED_MASK;
742 		newflags |= to_add;
743 
744 		/* Ignore errors */
745 		mlock_fixup(&vmi, vma, &prev, vma->vm_start, vma->vm_end,
746 			    newflags);
747 		cond_resched();
748 	}
749 out:
750 	return 0;
751 }
752 
753 SYSCALL_DEFINE1(mlockall, int, flags)
754 {
755 	unsigned long lock_limit;
756 	int ret;
757 
758 	if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)) ||
759 	    flags == MCL_ONFAULT)
760 		return -EINVAL;
761 
762 	if (!can_do_mlock())
763 		return -EPERM;
764 
765 	lock_limit = rlimit(RLIMIT_MEMLOCK);
766 	lock_limit >>= PAGE_SHIFT;
767 
768 	if (mmap_write_lock_killable(current->mm))
769 		return -EINTR;
770 
771 	ret = -ENOMEM;
772 	if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
773 	    capable(CAP_IPC_LOCK))
774 		ret = apply_mlockall_flags(flags);
775 	mmap_write_unlock(current->mm);
776 	if (!ret && (flags & MCL_CURRENT))
777 		mm_populate(0, TASK_SIZE);
778 
779 	return ret;
780 }
781 
782 SYSCALL_DEFINE0(munlockall)
783 {
784 	int ret;
785 
786 	if (mmap_write_lock_killable(current->mm))
787 		return -EINTR;
788 	ret = apply_mlockall_flags(0);
789 	mmap_write_unlock(current->mm);
790 	return ret;
791 }
792 
793 /*
794  * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
795  * shm segments) get accounted against the user_struct instead.
796  */
797 static DEFINE_SPINLOCK(shmlock_user_lock);
798 
799 int user_shm_lock(size_t size, struct ucounts *ucounts)
800 {
801 	unsigned long lock_limit, locked;
802 	long memlock;
803 	int allowed = 0;
804 
805 	locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
806 	lock_limit = rlimit(RLIMIT_MEMLOCK);
807 	if (lock_limit != RLIM_INFINITY)
808 		lock_limit >>= PAGE_SHIFT;
809 	spin_lock(&shmlock_user_lock);
810 	memlock = inc_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
811 
812 	if ((memlock == LONG_MAX || memlock > lock_limit) && !capable(CAP_IPC_LOCK)) {
813 		dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
814 		goto out;
815 	}
816 	if (!get_ucounts(ucounts)) {
817 		dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
818 		allowed = 0;
819 		goto out;
820 	}
821 	allowed = 1;
822 out:
823 	spin_unlock(&shmlock_user_lock);
824 	return allowed;
825 }
826 
827 void user_shm_unlock(size_t size, struct ucounts *ucounts)
828 {
829 	spin_lock(&shmlock_user_lock);
830 	dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, (size + PAGE_SIZE - 1) >> PAGE_SHIFT);
831 	spin_unlock(&shmlock_user_lock);
832 	put_ucounts(ucounts);
833 }
834