xref: /linux/mm/mlock.c (revision 79997eda0d31bc68203c95ecb978773ee6ce7a1f)
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->folios, folio_batch_count(fbatch));
210 	folio_batch_reinit(fbatch);
211 }
212 
213 void mlock_drain_local(void)
214 {
215 	struct folio_batch *fbatch;
216 
217 	local_lock(&mlock_fbatch.lock);
218 	fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
219 	if (folio_batch_count(fbatch))
220 		mlock_folio_batch(fbatch);
221 	local_unlock(&mlock_fbatch.lock);
222 }
223 
224 void mlock_drain_remote(int cpu)
225 {
226 	struct folio_batch *fbatch;
227 
228 	WARN_ON_ONCE(cpu_online(cpu));
229 	fbatch = &per_cpu(mlock_fbatch.fbatch, cpu);
230 	if (folio_batch_count(fbatch))
231 		mlock_folio_batch(fbatch);
232 }
233 
234 bool need_mlock_drain(int cpu)
235 {
236 	return folio_batch_count(&per_cpu(mlock_fbatch.fbatch, cpu));
237 }
238 
239 /**
240  * mlock_folio - mlock a folio already on (or temporarily off) LRU
241  * @folio: folio to be mlocked.
242  */
243 void mlock_folio(struct folio *folio)
244 {
245 	struct folio_batch *fbatch;
246 
247 	local_lock(&mlock_fbatch.lock);
248 	fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
249 
250 	if (!folio_test_set_mlocked(folio)) {
251 		int nr_pages = folio_nr_pages(folio);
252 
253 		zone_stat_mod_folio(folio, NR_MLOCK, nr_pages);
254 		__count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
255 	}
256 
257 	folio_get(folio);
258 	if (!folio_batch_add(fbatch, mlock_lru(folio)) ||
259 	    folio_test_large(folio) || lru_cache_disabled())
260 		mlock_folio_batch(fbatch);
261 	local_unlock(&mlock_fbatch.lock);
262 }
263 
264 /**
265  * mlock_new_folio - mlock a newly allocated folio not yet on LRU
266  * @folio: folio to be mlocked, either normal or a THP head.
267  */
268 void mlock_new_folio(struct folio *folio)
269 {
270 	struct folio_batch *fbatch;
271 	int nr_pages = folio_nr_pages(folio);
272 
273 	local_lock(&mlock_fbatch.lock);
274 	fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
275 	folio_set_mlocked(folio);
276 
277 	zone_stat_mod_folio(folio, NR_MLOCK, nr_pages);
278 	__count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
279 
280 	folio_get(folio);
281 	if (!folio_batch_add(fbatch, mlock_new(folio)) ||
282 	    folio_test_large(folio) || lru_cache_disabled())
283 		mlock_folio_batch(fbatch);
284 	local_unlock(&mlock_fbatch.lock);
285 }
286 
287 /**
288  * munlock_folio - munlock a folio
289  * @folio: folio to be munlocked, either normal or a THP head.
290  */
291 void munlock_folio(struct folio *folio)
292 {
293 	struct folio_batch *fbatch;
294 
295 	local_lock(&mlock_fbatch.lock);
296 	fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
297 	/*
298 	 * folio_test_clear_mlocked(folio) must be left to __munlock_folio(),
299 	 * which will check whether the folio is multiply mlocked.
300 	 */
301 	folio_get(folio);
302 	if (!folio_batch_add(fbatch, folio) ||
303 	    folio_test_large(folio) || lru_cache_disabled())
304 		mlock_folio_batch(fbatch);
305 	local_unlock(&mlock_fbatch.lock);
306 }
307 
308 static inline unsigned int folio_mlock_step(struct folio *folio,
309 		pte_t *pte, unsigned long addr, unsigned long end)
310 {
311 	unsigned int count, i, nr = folio_nr_pages(folio);
312 	unsigned long pfn = folio_pfn(folio);
313 	pte_t ptent = ptep_get(pte);
314 
315 	if (!folio_test_large(folio))
316 		return 1;
317 
318 	count = pfn + nr - pte_pfn(ptent);
319 	count = min_t(unsigned int, count, (end - addr) >> PAGE_SHIFT);
320 
321 	for (i = 0; i < count; i++, pte++) {
322 		pte_t entry = ptep_get(pte);
323 
324 		if (!pte_present(entry))
325 			break;
326 		if (pte_pfn(entry) - pfn >= nr)
327 			break;
328 	}
329 
330 	return i;
331 }
332 
333 static inline bool allow_mlock_munlock(struct folio *folio,
334 		struct vm_area_struct *vma, unsigned long start,
335 		unsigned long end, unsigned int step)
336 {
337 	/*
338 	 * For unlock, allow munlock large folio which is partially
339 	 * mapped to VMA. As it's possible that large folio is
340 	 * mlocked and VMA is split later.
341 	 *
342 	 * During memory pressure, such kind of large folio can
343 	 * be split. And the pages are not in VM_LOCKed VMA
344 	 * can be reclaimed.
345 	 */
346 	if (!(vma->vm_flags & VM_LOCKED))
347 		return true;
348 
349 	/* folio_within_range() cannot take KSM, but any small folio is OK */
350 	if (!folio_test_large(folio))
351 		return true;
352 
353 	/* folio not in range [start, end), skip mlock */
354 	if (!folio_within_range(folio, vma, start, end))
355 		return false;
356 
357 	/* folio is not fully mapped, skip mlock */
358 	if (step != folio_nr_pages(folio))
359 		return false;
360 
361 	return true;
362 }
363 
364 static int mlock_pte_range(pmd_t *pmd, unsigned long addr,
365 			   unsigned long end, struct mm_walk *walk)
366 
367 {
368 	struct vm_area_struct *vma = walk->vma;
369 	spinlock_t *ptl;
370 	pte_t *start_pte, *pte;
371 	pte_t ptent;
372 	struct folio *folio;
373 	unsigned int step = 1;
374 	unsigned long start = addr;
375 
376 	ptl = pmd_trans_huge_lock(pmd, vma);
377 	if (ptl) {
378 		if (!pmd_present(*pmd))
379 			goto out;
380 		if (is_huge_zero_pmd(*pmd))
381 			goto out;
382 		folio = page_folio(pmd_page(*pmd));
383 		if (vma->vm_flags & VM_LOCKED)
384 			mlock_folio(folio);
385 		else
386 			munlock_folio(folio);
387 		goto out;
388 	}
389 
390 	start_pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
391 	if (!start_pte) {
392 		walk->action = ACTION_AGAIN;
393 		return 0;
394 	}
395 
396 	for (pte = start_pte; addr != end; pte++, addr += PAGE_SIZE) {
397 		ptent = ptep_get(pte);
398 		if (!pte_present(ptent))
399 			continue;
400 		folio = vm_normal_folio(vma, addr, ptent);
401 		if (!folio || folio_is_zone_device(folio))
402 			continue;
403 
404 		step = folio_mlock_step(folio, pte, addr, end);
405 		if (!allow_mlock_munlock(folio, vma, start, end, step))
406 			goto next_entry;
407 
408 		if (vma->vm_flags & VM_LOCKED)
409 			mlock_folio(folio);
410 		else
411 			munlock_folio(folio);
412 
413 next_entry:
414 		pte += step - 1;
415 		addr += (step - 1) << PAGE_SHIFT;
416 	}
417 	pte_unmap(start_pte);
418 out:
419 	spin_unlock(ptl);
420 	cond_resched();
421 	return 0;
422 }
423 
424 /*
425  * mlock_vma_pages_range() - mlock any pages already in the range,
426  *                           or munlock all pages in the range.
427  * @vma - vma containing range to be mlock()ed or munlock()ed
428  * @start - start address in @vma of the range
429  * @end - end of range in @vma
430  * @newflags - the new set of flags for @vma.
431  *
432  * Called for mlock(), mlock2() and mlockall(), to set @vma VM_LOCKED;
433  * called for munlock() and munlockall(), to clear VM_LOCKED from @vma.
434  */
435 static void mlock_vma_pages_range(struct vm_area_struct *vma,
436 	unsigned long start, unsigned long end, vm_flags_t newflags)
437 {
438 	static const struct mm_walk_ops mlock_walk_ops = {
439 		.pmd_entry = mlock_pte_range,
440 		.walk_lock = PGWALK_WRLOCK_VERIFY,
441 	};
442 
443 	/*
444 	 * There is a slight chance that concurrent page migration,
445 	 * or page reclaim finding a page of this now-VM_LOCKED vma,
446 	 * will call mlock_vma_folio() and raise page's mlock_count:
447 	 * double counting, leaving the page unevictable indefinitely.
448 	 * Communicate this danger to mlock_vma_folio() with VM_IO,
449 	 * which is a VM_SPECIAL flag not allowed on VM_LOCKED vmas.
450 	 * mmap_lock is held in write mode here, so this weird
451 	 * combination should not be visible to other mmap_lock users;
452 	 * but WRITE_ONCE so rmap walkers must see VM_IO if VM_LOCKED.
453 	 */
454 	if (newflags & VM_LOCKED)
455 		newflags |= VM_IO;
456 	vma_start_write(vma);
457 	vm_flags_reset_once(vma, newflags);
458 
459 	lru_add_drain();
460 	walk_page_range(vma->vm_mm, start, end, &mlock_walk_ops, NULL);
461 	lru_add_drain();
462 
463 	if (newflags & VM_IO) {
464 		newflags &= ~VM_IO;
465 		vm_flags_reset_once(vma, newflags);
466 	}
467 }
468 
469 /*
470  * mlock_fixup  - handle mlock[all]/munlock[all] requests.
471  *
472  * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
473  * munlock is a no-op.  However, for some special vmas, we go ahead and
474  * populate the ptes.
475  *
476  * For vmas that pass the filters, merge/split as appropriate.
477  */
478 static int mlock_fixup(struct vma_iterator *vmi, struct vm_area_struct *vma,
479 	       struct vm_area_struct **prev, unsigned long start,
480 	       unsigned long end, vm_flags_t newflags)
481 {
482 	struct mm_struct *mm = vma->vm_mm;
483 	int nr_pages;
484 	int ret = 0;
485 	vm_flags_t oldflags = vma->vm_flags;
486 
487 	if (newflags == oldflags || (oldflags & VM_SPECIAL) ||
488 	    is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm) ||
489 	    vma_is_dax(vma) || vma_is_secretmem(vma))
490 		/* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
491 		goto out;
492 
493 	vma = vma_modify_flags(vmi, *prev, vma, start, end, newflags);
494 	if (IS_ERR(vma)) {
495 		ret = PTR_ERR(vma);
496 		goto out;
497 	}
498 
499 	/*
500 	 * Keep track of amount of locked VM.
501 	 */
502 	nr_pages = (end - start) >> PAGE_SHIFT;
503 	if (!(newflags & VM_LOCKED))
504 		nr_pages = -nr_pages;
505 	else if (oldflags & VM_LOCKED)
506 		nr_pages = 0;
507 	mm->locked_vm += nr_pages;
508 
509 	/*
510 	 * vm_flags is protected by the mmap_lock held in write mode.
511 	 * It's okay if try_to_unmap_one unmaps a page just after we
512 	 * set VM_LOCKED, populate_vma_page_range will bring it back.
513 	 */
514 	if ((newflags & VM_LOCKED) && (oldflags & VM_LOCKED)) {
515 		/* No work to do, and mlocking twice would be wrong */
516 		vma_start_write(vma);
517 		vm_flags_reset(vma, newflags);
518 	} else {
519 		mlock_vma_pages_range(vma, start, end, newflags);
520 	}
521 out:
522 	*prev = vma;
523 	return ret;
524 }
525 
526 static int apply_vma_lock_flags(unsigned long start, size_t len,
527 				vm_flags_t flags)
528 {
529 	unsigned long nstart, end, tmp;
530 	struct vm_area_struct *vma, *prev;
531 	VMA_ITERATOR(vmi, current->mm, start);
532 
533 	VM_BUG_ON(offset_in_page(start));
534 	VM_BUG_ON(len != PAGE_ALIGN(len));
535 	end = start + len;
536 	if (end < start)
537 		return -EINVAL;
538 	if (end == start)
539 		return 0;
540 	vma = vma_iter_load(&vmi);
541 	if (!vma)
542 		return -ENOMEM;
543 
544 	prev = vma_prev(&vmi);
545 	if (start > vma->vm_start)
546 		prev = vma;
547 
548 	nstart = start;
549 	tmp = vma->vm_start;
550 	for_each_vma_range(vmi, vma, end) {
551 		int error;
552 		vm_flags_t newflags;
553 
554 		if (vma->vm_start != tmp)
555 			return -ENOMEM;
556 
557 		newflags = vma->vm_flags & ~VM_LOCKED_MASK;
558 		newflags |= flags;
559 		/* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
560 		tmp = vma->vm_end;
561 		if (tmp > end)
562 			tmp = end;
563 		error = mlock_fixup(&vmi, vma, &prev, nstart, tmp, newflags);
564 		if (error)
565 			return error;
566 		tmp = vma_iter_end(&vmi);
567 		nstart = tmp;
568 	}
569 
570 	if (tmp < end)
571 		return -ENOMEM;
572 
573 	return 0;
574 }
575 
576 /*
577  * Go through vma areas and sum size of mlocked
578  * vma pages, as return value.
579  * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT)
580  * is also counted.
581  * Return value: previously mlocked page counts
582  */
583 static unsigned long count_mm_mlocked_page_nr(struct mm_struct *mm,
584 		unsigned long start, size_t len)
585 {
586 	struct vm_area_struct *vma;
587 	unsigned long count = 0;
588 	unsigned long end;
589 	VMA_ITERATOR(vmi, mm, start);
590 
591 	/* Don't overflow past ULONG_MAX */
592 	if (unlikely(ULONG_MAX - len < start))
593 		end = ULONG_MAX;
594 	else
595 		end = start + len;
596 
597 	for_each_vma_range(vmi, vma, end) {
598 		if (vma->vm_flags & VM_LOCKED) {
599 			if (start > vma->vm_start)
600 				count -= (start - vma->vm_start);
601 			if (end < vma->vm_end) {
602 				count += end - vma->vm_start;
603 				break;
604 			}
605 			count += vma->vm_end - vma->vm_start;
606 		}
607 	}
608 
609 	return count >> PAGE_SHIFT;
610 }
611 
612 /*
613  * convert get_user_pages() return value to posix mlock() error
614  */
615 static int __mlock_posix_error_return(long retval)
616 {
617 	if (retval == -EFAULT)
618 		retval = -ENOMEM;
619 	else if (retval == -ENOMEM)
620 		retval = -EAGAIN;
621 	return retval;
622 }
623 
624 static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
625 {
626 	unsigned long locked;
627 	unsigned long lock_limit;
628 	int error = -ENOMEM;
629 
630 	start = untagged_addr(start);
631 
632 	if (!can_do_mlock())
633 		return -EPERM;
634 
635 	len = PAGE_ALIGN(len + (offset_in_page(start)));
636 	start &= PAGE_MASK;
637 
638 	lock_limit = rlimit(RLIMIT_MEMLOCK);
639 	lock_limit >>= PAGE_SHIFT;
640 	locked = len >> PAGE_SHIFT;
641 
642 	if (mmap_write_lock_killable(current->mm))
643 		return -EINTR;
644 
645 	locked += current->mm->locked_vm;
646 	if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) {
647 		/*
648 		 * It is possible that the regions requested intersect with
649 		 * previously mlocked areas, that part area in "mm->locked_vm"
650 		 * should not be counted to new mlock increment count. So check
651 		 * and adjust locked count if necessary.
652 		 */
653 		locked -= count_mm_mlocked_page_nr(current->mm,
654 				start, len);
655 	}
656 
657 	/* check against resource limits */
658 	if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
659 		error = apply_vma_lock_flags(start, len, flags);
660 
661 	mmap_write_unlock(current->mm);
662 	if (error)
663 		return error;
664 
665 	error = __mm_populate(start, len, 0);
666 	if (error)
667 		return __mlock_posix_error_return(error);
668 	return 0;
669 }
670 
671 SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
672 {
673 	return do_mlock(start, len, VM_LOCKED);
674 }
675 
676 SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
677 {
678 	vm_flags_t vm_flags = VM_LOCKED;
679 
680 	if (flags & ~MLOCK_ONFAULT)
681 		return -EINVAL;
682 
683 	if (flags & MLOCK_ONFAULT)
684 		vm_flags |= VM_LOCKONFAULT;
685 
686 	return do_mlock(start, len, vm_flags);
687 }
688 
689 SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
690 {
691 	int ret;
692 
693 	start = untagged_addr(start);
694 
695 	len = PAGE_ALIGN(len + (offset_in_page(start)));
696 	start &= PAGE_MASK;
697 
698 	if (mmap_write_lock_killable(current->mm))
699 		return -EINTR;
700 	ret = apply_vma_lock_flags(start, len, 0);
701 	mmap_write_unlock(current->mm);
702 
703 	return ret;
704 }
705 
706 /*
707  * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
708  * and translate into the appropriate modifications to mm->def_flags and/or the
709  * flags for all current VMAs.
710  *
711  * There are a couple of subtleties with this.  If mlockall() is called multiple
712  * times with different flags, the values do not necessarily stack.  If mlockall
713  * is called once including the MCL_FUTURE flag and then a second time without
714  * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
715  */
716 static int apply_mlockall_flags(int flags)
717 {
718 	VMA_ITERATOR(vmi, current->mm, 0);
719 	struct vm_area_struct *vma, *prev = NULL;
720 	vm_flags_t to_add = 0;
721 
722 	current->mm->def_flags &= ~VM_LOCKED_MASK;
723 	if (flags & MCL_FUTURE) {
724 		current->mm->def_flags |= VM_LOCKED;
725 
726 		if (flags & MCL_ONFAULT)
727 			current->mm->def_flags |= VM_LOCKONFAULT;
728 
729 		if (!(flags & MCL_CURRENT))
730 			goto out;
731 	}
732 
733 	if (flags & MCL_CURRENT) {
734 		to_add |= VM_LOCKED;
735 		if (flags & MCL_ONFAULT)
736 			to_add |= VM_LOCKONFAULT;
737 	}
738 
739 	for_each_vma(vmi, vma) {
740 		vm_flags_t newflags;
741 
742 		newflags = vma->vm_flags & ~VM_LOCKED_MASK;
743 		newflags |= to_add;
744 
745 		/* Ignore errors */
746 		mlock_fixup(&vmi, vma, &prev, vma->vm_start, vma->vm_end,
747 			    newflags);
748 		cond_resched();
749 	}
750 out:
751 	return 0;
752 }
753 
754 SYSCALL_DEFINE1(mlockall, int, flags)
755 {
756 	unsigned long lock_limit;
757 	int ret;
758 
759 	if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)) ||
760 	    flags == MCL_ONFAULT)
761 		return -EINVAL;
762 
763 	if (!can_do_mlock())
764 		return -EPERM;
765 
766 	lock_limit = rlimit(RLIMIT_MEMLOCK);
767 	lock_limit >>= PAGE_SHIFT;
768 
769 	if (mmap_write_lock_killable(current->mm))
770 		return -EINTR;
771 
772 	ret = -ENOMEM;
773 	if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
774 	    capable(CAP_IPC_LOCK))
775 		ret = apply_mlockall_flags(flags);
776 	mmap_write_unlock(current->mm);
777 	if (!ret && (flags & MCL_CURRENT))
778 		mm_populate(0, TASK_SIZE);
779 
780 	return ret;
781 }
782 
783 SYSCALL_DEFINE0(munlockall)
784 {
785 	int ret;
786 
787 	if (mmap_write_lock_killable(current->mm))
788 		return -EINTR;
789 	ret = apply_mlockall_flags(0);
790 	mmap_write_unlock(current->mm);
791 	return ret;
792 }
793 
794 /*
795  * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
796  * shm segments) get accounted against the user_struct instead.
797  */
798 static DEFINE_SPINLOCK(shmlock_user_lock);
799 
800 int user_shm_lock(size_t size, struct ucounts *ucounts)
801 {
802 	unsigned long lock_limit, locked;
803 	long memlock;
804 	int allowed = 0;
805 
806 	locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
807 	lock_limit = rlimit(RLIMIT_MEMLOCK);
808 	if (lock_limit != RLIM_INFINITY)
809 		lock_limit >>= PAGE_SHIFT;
810 	spin_lock(&shmlock_user_lock);
811 	memlock = inc_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
812 
813 	if ((memlock == LONG_MAX || memlock > lock_limit) && !capable(CAP_IPC_LOCK)) {
814 		dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
815 		goto out;
816 	}
817 	if (!get_ucounts(ucounts)) {
818 		dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
819 		allowed = 0;
820 		goto out;
821 	}
822 	allowed = 1;
823 out:
824 	spin_unlock(&shmlock_user_lock);
825 	return allowed;
826 }
827 
828 void user_shm_unlock(size_t size, struct ucounts *ucounts)
829 {
830 	spin_lock(&shmlock_user_lock);
831 	dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, (size + PAGE_SIZE - 1) >> PAGE_SHIFT);
832 	spin_unlock(&shmlock_user_lock);
833 	put_ucounts(ucounts);
834 }
835