Linux Audio

Check our new training course

Loading...
v6.2
  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
 31struct mlock_pvec {
 32	local_lock_t lock;
 33	struct pagevec vec;
 34};
 35
 36static DEFINE_PER_CPU(struct mlock_pvec, mlock_pvec) = {
 37	.lock = INIT_LOCAL_LOCK(lock),
 38};
 39
 40bool 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}
 48EXPORT_SYMBOL(can_do_mlock);
 49
 50/*
 51 * Mlocked pages are marked with PageMlocked() flag for efficient testing
 52 * in vmscan and, possibly, the fault path; and to support semi-accurate
 53 * statistics.
 54 *
 55 * An mlocked page [PageMlocked(page)] is unevictable.  As such, it will
 56 * be placed on the LRU "unevictable" list, rather than the [in]active lists.
 57 * The unevictable list is an LRU sibling list to the [in]active lists.
 58 * PageUnevictable is set to indicate the unevictable state.
 
 
 
 
 
 
 59 */
 60
 61static struct lruvec *__mlock_page(struct page *page, struct lruvec *lruvec)
 
 
 
 62{
 63	/* There is nothing more we can do while it's off LRU */
 64	if (!TestClearPageLRU(page))
 65		return lruvec;
 66
 67	lruvec = folio_lruvec_relock_irq(page_folio(page), lruvec);
 
 68
 69	if (unlikely(page_evictable(page))) {
 
 
 
 
 
 
 
 
 
 
 
 70		/*
 71		 * This is a little surprising, but quite possible:
 72		 * PageMlocked must have got cleared already by another CPU.
 73		 * Could this page be on the Unevictable LRU?  I'm not sure,
 74		 * but move it now if so.
 75		 */
 76		if (PageUnevictable(page)) {
 77			del_page_from_lru_list(page, lruvec);
 78			ClearPageUnevictable(page);
 79			add_page_to_lru_list(page, lruvec);
 80			__count_vm_events(UNEVICTABLE_PGRESCUED,
 81					  thp_nr_pages(page));
 82		}
 83		goto out;
 84	}
 85
 86	if (PageUnevictable(page)) {
 87		if (PageMlocked(page))
 88			page->mlock_count++;
 89		goto out;
 90	}
 91
 92	del_page_from_lru_list(page, lruvec);
 93	ClearPageActive(page);
 94	SetPageUnevictable(page);
 95	page->mlock_count = !!PageMlocked(page);
 96	add_page_to_lru_list(page, lruvec);
 97	__count_vm_events(UNEVICTABLE_PGCULLED, thp_nr_pages(page));
 98out:
 99	SetPageLRU(page);
100	return lruvec;
101}
102
103static struct lruvec *__mlock_new_page(struct page *page, struct lruvec *lruvec)
 
 
 
 
104{
105	VM_BUG_ON_PAGE(PageLRU(page), page);
 
106
107	lruvec = folio_lruvec_relock_irq(page_folio(page), lruvec);
 
108
109	/* As above, this is a little surprising, but possible */
110	if (unlikely(page_evictable(page)))
111		goto out;
112
113	SetPageUnevictable(page);
114	page->mlock_count = !!PageMlocked(page);
115	__count_vm_events(UNEVICTABLE_PGCULLED, thp_nr_pages(page));
116out:
117	add_page_to_lru_list(page, lruvec);
118	SetPageLRU(page);
119	return lruvec;
120}
121
122static struct lruvec *__munlock_page(struct page *page, struct lruvec *lruvec)
 
 
 
 
123{
124	int nr_pages = thp_nr_pages(page);
125	bool isolated = false;
126
127	if (!TestClearPageLRU(page))
128		goto munlock;
129
130	isolated = true;
131	lruvec = folio_lruvec_relock_irq(page_folio(page), lruvec);
132
133	if (PageUnevictable(page)) {
134		/* Then mlock_count is maintained, but might undercount */
135		if (page->mlock_count)
136			page->mlock_count--;
137		if (page->mlock_count)
138			goto out;
139	}
140	/* else assume that was the last mlock: reclaim will fix it if not */
141
142munlock:
143	if (TestClearPageMlocked(page)) {
144		__mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
145		if (isolated || !PageUnevictable(page))
146			__count_vm_events(UNEVICTABLE_PGMUNLOCKED, nr_pages);
147		else
148			__count_vm_events(UNEVICTABLE_PGSTRANDED, nr_pages);
149	}
150
151	/* page_evictable() has to be checked *after* clearing Mlocked */
152	if (isolated && PageUnevictable(page) && page_evictable(page)) {
153		del_page_from_lru_list(page, lruvec);
154		ClearPageUnevictable(page);
155		add_page_to_lru_list(page, lruvec);
156		__count_vm_events(UNEVICTABLE_PGRESCUED, nr_pages);
157	}
158out:
159	if (isolated)
160		SetPageLRU(page);
161	return lruvec;
162}
163
164/*
165 * Flags held in the low bits of a struct page pointer on the mlock_pvec.
 
 
 
166 */
167#define LRU_PAGE 0x1
168#define NEW_PAGE 0x2
169static inline struct page *mlock_lru(struct page *page)
170{
171	return (struct page *)((unsigned long)page + LRU_PAGE);
 
 
 
 
 
 
 
 
 
 
 
172}
173
174static inline struct page *mlock_new(struct page *page)
 
 
 
 
 
 
 
 
 
175{
176	return (struct page *)((unsigned long)page + NEW_PAGE);
 
 
 
 
 
177}
178
179/*
180 * mlock_pagevec() is derived from pagevec_lru_move_fn():
181 * perhaps that can make use of such page pointer flags in future,
182 * but for now just keep it for mlock.  We could use three separate
183 * pagevecs instead, but one feels better (munlocking a full pagevec
184 * does not need to drain mlocking pagevecs first).
 
 
 
 
 
 
 
 
 
 
 
185 */
186static void mlock_pagevec(struct pagevec *pvec)
187{
188	struct lruvec *lruvec = NULL;
189	unsigned long mlock;
190	struct page *page;
191	int i;
192
193	for (i = 0; i < pagevec_count(pvec); i++) {
194		page = pvec->pages[i];
195		mlock = (unsigned long)page & (LRU_PAGE | NEW_PAGE);
196		page = (struct page *)((unsigned long)page - mlock);
197		pvec->pages[i] = page;
198
199		if (mlock & LRU_PAGE)
200			lruvec = __mlock_page(page, lruvec);
201		else if (mlock & NEW_PAGE)
202			lruvec = __mlock_new_page(page, lruvec);
203		else
204			lruvec = __munlock_page(page, lruvec);
205	}
206
207	if (lruvec)
208		unlock_page_lruvec_irq(lruvec);
209	release_pages(pvec->pages, pvec->nr);
210	pagevec_reinit(pvec);
211}
 
212
213void mlock_page_drain_local(void)
214{
215	struct pagevec *pvec;
 
 
216
217	local_lock(&mlock_pvec.lock);
218	pvec = this_cpu_ptr(&mlock_pvec.vec);
219	if (pagevec_count(pvec))
220		mlock_pagevec(pvec);
221	local_unlock(&mlock_pvec.lock);
222}
223
224void mlock_page_drain_remote(int cpu)
225{
226	struct pagevec *pvec;
 
 
 
227
228	WARN_ON_ONCE(cpu_online(cpu));
229	pvec = &per_cpu(mlock_pvec.vec, cpu);
230	if (pagevec_count(pvec))
231		mlock_pagevec(pvec);
 
232}
233
234bool need_mlock_page_drain(int cpu)
 
 
 
235{
236	return pagevec_count(&per_cpu(mlock_pvec.vec, cpu));
 
 
 
 
237}
238
239/**
240 * mlock_folio - mlock a folio already on (or temporarily off) LRU
241 * @folio: folio to be mlocked.
 
 
 
 
 
 
 
 
242 */
243void mlock_folio(struct folio *folio)
 
244{
245	struct pagevec *pvec;
246
247	local_lock(&mlock_pvec.lock);
248	pvec = this_cpu_ptr(&mlock_pvec.vec);
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 (!pagevec_add(pvec, mlock_lru(&folio->page)) ||
259	    folio_test_large(folio) || lru_cache_disabled())
260		mlock_pagevec(pvec);
261	local_unlock(&mlock_pvec.lock);
262}
263
264/**
265 * mlock_new_page - mlock a newly allocated page not yet on LRU
266 * @page: page to be mlocked, either a normal page or a THP head.
 
 
267 */
268void mlock_new_page(struct page *page)
269{
270	struct pagevec *pvec;
271	int nr_pages = thp_nr_pages(page);
272
273	local_lock(&mlock_pvec.lock);
274	pvec = this_cpu_ptr(&mlock_pvec.vec);
275	SetPageMlocked(page);
276	mod_zone_page_state(page_zone(page), NR_MLOCK, nr_pages);
277	__count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
278
279	get_page(page);
280	if (!pagevec_add(pvec, mlock_new(page)) ||
281	    PageHead(page) || lru_cache_disabled())
282		mlock_pagevec(pvec);
283	local_unlock(&mlock_pvec.lock);
284}
285
286/**
287 * munlock_page - munlock a page
288 * @page: page to be munlocked, either a normal page or a THP head.
 
 
 
 
 
 
289 */
290void munlock_page(struct page *page)
291{
292	struct pagevec *pvec;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
293
294	local_lock(&mlock_pvec.lock);
295	pvec = this_cpu_ptr(&mlock_pvec.vec);
296	/*
297	 * TestClearPageMlocked(page) must be left to __munlock_page(),
298	 * which will check whether the page is multiply mlocked.
299	 */
300
301	get_page(page);
302	if (!pagevec_add(pvec, page) ||
303	    PageHead(page) || lru_cache_disabled())
304		mlock_pagevec(pvec);
305	local_unlock(&mlock_pvec.lock);
306}
307
308static int mlock_pte_range(pmd_t *pmd, unsigned long addr,
309			   unsigned long end, struct mm_walk *walk)
310
 
 
 
 
 
 
 
 
 
 
 
 
 
311{
312	struct vm_area_struct *vma = walk->vma;
313	spinlock_t *ptl;
314	pte_t *start_pte, *pte;
315	struct page *page;
316
317	ptl = pmd_trans_huge_lock(pmd, vma);
318	if (ptl) {
319		if (!pmd_present(*pmd))
320			goto out;
321		if (is_huge_zero_pmd(*pmd))
322			goto out;
323		page = pmd_page(*pmd);
324		if (vma->vm_flags & VM_LOCKED)
325			mlock_folio(page_folio(page));
326		else
327			munlock_page(page);
328		goto out;
329	}
 
 
 
 
 
 
 
 
 
 
 
 
330
331	start_pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
332	for (pte = start_pte; addr != end; pte++, addr += PAGE_SIZE) {
333		if (!pte_present(*pte))
334			continue;
335		page = vm_normal_page(vma, addr, *pte);
336		if (!page || is_zone_device_page(page))
337			continue;
338		if (PageTransCompound(page))
339			continue;
340		if (vma->vm_flags & VM_LOCKED)
341			mlock_folio(page_folio(page));
342		else
343			munlock_page(page);
 
 
 
 
 
344	}
345	pte_unmap(start_pte);
346out:
347	spin_unlock(ptl);
348	cond_resched();
349	return 0;
350}
351
352/*
353 * mlock_vma_pages_range() - mlock any pages already in the range,
354 *                           or munlock all pages in the range.
355 * @vma - vma containing range to be mlock()ed or munlock()ed
356 * @start - start address in @vma of the range
357 * @end - end of range in @vma
358 * @newflags - the new set of flags for @vma.
359 *
360 * Called for mlock(), mlock2() and mlockall(), to set @vma VM_LOCKED;
361 * called for munlock() and munlockall(), to clear VM_LOCKED from @vma.
362 */
363static void mlock_vma_pages_range(struct vm_area_struct *vma,
364	unsigned long start, unsigned long end, vm_flags_t newflags)
365{
366	static const struct mm_walk_ops mlock_walk_ops = {
367		.pmd_entry = mlock_pte_range,
368	};
 
 
 
 
 
 
 
 
 
 
 
 
 
 
369
370	/*
371	 * There is a slight chance that concurrent page migration,
372	 * or page reclaim finding a page of this now-VM_LOCKED vma,
373	 * will call mlock_vma_page() and raise page's mlock_count:
374	 * double counting, leaving the page unevictable indefinitely.
375	 * Communicate this danger to mlock_vma_page() with VM_IO,
376	 * which is a VM_SPECIAL flag not allowed on VM_LOCKED vmas.
377	 * mmap_lock is held in write mode here, so this weird
378	 * combination should not be visible to other mmap_lock users;
379	 * but WRITE_ONCE so rmap walkers must see VM_IO if VM_LOCKED.
380	 */
381	if (newflags & VM_LOCKED)
382		newflags |= VM_IO;
383	WRITE_ONCE(vma->vm_flags, newflags);
384
385	lru_add_drain();
386	walk_page_range(vma->vm_mm, start, end, &mlock_walk_ops, NULL);
387	lru_add_drain();
388
389	if (newflags & VM_IO) {
390		newflags &= ~VM_IO;
391		WRITE_ONCE(vma->vm_flags, newflags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
392	}
393}
394
395/*
396 * mlock_fixup  - handle mlock[all]/munlock[all] requests.
397 *
398 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
399 * munlock is a no-op.  However, for some special vmas, we go ahead and
400 * populate the ptes.
401 *
402 * For vmas that pass the filters, merge/split as appropriate.
403 */
404static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
405	unsigned long start, unsigned long end, vm_flags_t newflags)
406{
407	struct mm_struct *mm = vma->vm_mm;
408	pgoff_t pgoff;
409	int nr_pages;
410	int ret = 0;
411	vm_flags_t oldflags = vma->vm_flags;
 
412
413	if (newflags == oldflags || (oldflags & VM_SPECIAL) ||
414	    is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm) ||
415	    vma_is_dax(vma) || vma_is_secretmem(vma))
416		/* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
417		goto out;
418
419	pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
420	*prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
421			  vma->vm_file, pgoff, vma_policy(vma),
422			  vma->vm_userfaultfd_ctx, anon_vma_name(vma));
423	if (*prev) {
424		vma = *prev;
425		goto success;
426	}
427
428	if (start != vma->vm_start) {
429		ret = split_vma(mm, vma, start, 1);
430		if (ret)
431			goto out;
432	}
433
434	if (end != vma->vm_end) {
435		ret = split_vma(mm, vma, end, 0);
436		if (ret)
437			goto out;
438	}
439
440success:
441	/*
442	 * Keep track of amount of locked VM.
443	 */
444	nr_pages = (end - start) >> PAGE_SHIFT;
445	if (!(newflags & VM_LOCKED))
446		nr_pages = -nr_pages;
447	else if (oldflags & VM_LOCKED)
448		nr_pages = 0;
449	mm->locked_vm += nr_pages;
450
451	/*
452	 * vm_flags is protected by the mmap_lock held in write mode.
453	 * It's okay if try_to_unmap_one unmaps a page just after we
454	 * set VM_LOCKED, populate_vma_page_range will bring it back.
455	 */
456
457	if ((newflags & VM_LOCKED) && (oldflags & VM_LOCKED)) {
458		/* No work to do, and mlocking twice would be wrong */
459		vma->vm_flags = newflags;
460	} else {
461		mlock_vma_pages_range(vma, start, end, newflags);
462	}
463out:
464	*prev = vma;
465	return ret;
466}
467
468static int apply_vma_lock_flags(unsigned long start, size_t len,
469				vm_flags_t flags)
470{
471	unsigned long nstart, end, tmp;
472	struct vm_area_struct *vma, *prev;
473	int error;
474	MA_STATE(mas, &current->mm->mm_mt, start, start);
475
476	VM_BUG_ON(offset_in_page(start));
477	VM_BUG_ON(len != PAGE_ALIGN(len));
478	end = start + len;
479	if (end < start)
480		return -EINVAL;
481	if (end == start)
482		return 0;
483	vma = mas_walk(&mas);
484	if (!vma)
485		return -ENOMEM;
486
 
487	if (start > vma->vm_start)
488		prev = vma;
489	else
490		prev = mas_prev(&mas, 0);
491
492	for (nstart = start ; ; ) {
493		vm_flags_t newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
494
495		newflags |= flags;
496
497		/* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
498		tmp = vma->vm_end;
499		if (tmp > end)
500			tmp = end;
501		error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
502		if (error)
503			break;
504		nstart = tmp;
505		if (nstart < prev->vm_end)
506			nstart = prev->vm_end;
507		if (nstart >= end)
508			break;
509
510		vma = find_vma(prev->vm_mm, prev->vm_end);
511		if (!vma || vma->vm_start != nstart) {
512			error = -ENOMEM;
513			break;
514		}
515	}
516	return error;
517}
518
519/*
520 * Go through vma areas and sum size of mlocked
521 * vma pages, as return value.
522 * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT)
523 * is also counted.
524 * Return value: previously mlocked page counts
525 */
526static unsigned long count_mm_mlocked_page_nr(struct mm_struct *mm,
527		unsigned long start, size_t len)
528{
529	struct vm_area_struct *vma;
530	unsigned long count = 0;
531	unsigned long end;
532	VMA_ITERATOR(vmi, mm, start);
533
534	/* Don't overflow past ULONG_MAX */
535	if (unlikely(ULONG_MAX - len < start))
536		end = ULONG_MAX;
537	else
538		end = start + len;
 
539
540	for_each_vma_range(vmi, vma, end) {
 
 
 
 
541		if (vma->vm_flags & VM_LOCKED) {
542			if (start > vma->vm_start)
543				count -= (start - vma->vm_start);
544			if (end < vma->vm_end) {
545				count += end - vma->vm_start;
546				break;
547			}
548			count += vma->vm_end - vma->vm_start;
549		}
550	}
551
552	return count >> PAGE_SHIFT;
553}
554
555/*
556 * convert get_user_pages() return value to posix mlock() error
557 */
558static int __mlock_posix_error_return(long retval)
559{
560	if (retval == -EFAULT)
561		retval = -ENOMEM;
562	else if (retval == -ENOMEM)
563		retval = -EAGAIN;
564	return retval;
565}
566
567static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
568{
569	unsigned long locked;
570	unsigned long lock_limit;
571	int error = -ENOMEM;
572
573	start = untagged_addr(start);
574
575	if (!can_do_mlock())
576		return -EPERM;
577
578	len = PAGE_ALIGN(len + (offset_in_page(start)));
579	start &= PAGE_MASK;
580
581	lock_limit = rlimit(RLIMIT_MEMLOCK);
582	lock_limit >>= PAGE_SHIFT;
583	locked = len >> PAGE_SHIFT;
584
585	if (mmap_write_lock_killable(current->mm))
586		return -EINTR;
587
588	locked += current->mm->locked_vm;
589	if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) {
590		/*
591		 * It is possible that the regions requested intersect with
592		 * previously mlocked areas, that part area in "mm->locked_vm"
593		 * should not be counted to new mlock increment count. So check
594		 * and adjust locked count if necessary.
595		 */
596		locked -= count_mm_mlocked_page_nr(current->mm,
597				start, len);
598	}
599
600	/* check against resource limits */
601	if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
602		error = apply_vma_lock_flags(start, len, flags);
603
604	mmap_write_unlock(current->mm);
605	if (error)
606		return error;
607
608	error = __mm_populate(start, len, 0);
609	if (error)
610		return __mlock_posix_error_return(error);
611	return 0;
612}
613
614SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
615{
616	return do_mlock(start, len, VM_LOCKED);
617}
618
619SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
620{
621	vm_flags_t vm_flags = VM_LOCKED;
622
623	if (flags & ~MLOCK_ONFAULT)
624		return -EINVAL;
625
626	if (flags & MLOCK_ONFAULT)
627		vm_flags |= VM_LOCKONFAULT;
628
629	return do_mlock(start, len, vm_flags);
630}
631
632SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
633{
634	int ret;
635
636	start = untagged_addr(start);
637
638	len = PAGE_ALIGN(len + (offset_in_page(start)));
639	start &= PAGE_MASK;
640
641	if (mmap_write_lock_killable(current->mm))
642		return -EINTR;
643	ret = apply_vma_lock_flags(start, len, 0);
644	mmap_write_unlock(current->mm);
645
646	return ret;
647}
648
649/*
650 * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
651 * and translate into the appropriate modifications to mm->def_flags and/or the
652 * flags for all current VMAs.
653 *
654 * There are a couple of subtleties with this.  If mlockall() is called multiple
655 * times with different flags, the values do not necessarily stack.  If mlockall
656 * is called once including the MCL_FUTURE flag and then a second time without
657 * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
658 */
659static int apply_mlockall_flags(int flags)
660{
661	MA_STATE(mas, &current->mm->mm_mt, 0, 0);
662	struct vm_area_struct *vma, *prev = NULL;
663	vm_flags_t to_add = 0;
664
665	current->mm->def_flags &= VM_LOCKED_CLEAR_MASK;
666	if (flags & MCL_FUTURE) {
667		current->mm->def_flags |= VM_LOCKED;
668
669		if (flags & MCL_ONFAULT)
670			current->mm->def_flags |= VM_LOCKONFAULT;
671
672		if (!(flags & MCL_CURRENT))
673			goto out;
674	}
675
676	if (flags & MCL_CURRENT) {
677		to_add |= VM_LOCKED;
678		if (flags & MCL_ONFAULT)
679			to_add |= VM_LOCKONFAULT;
680	}
681
682	mas_for_each(&mas, vma, ULONG_MAX) {
683		vm_flags_t newflags;
684
685		newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
686		newflags |= to_add;
687
688		/* Ignore errors */
689		mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
690		mas_pause(&mas);
691		cond_resched();
692	}
693out:
694	return 0;
695}
696
697SYSCALL_DEFINE1(mlockall, int, flags)
698{
699	unsigned long lock_limit;
700	int ret;
701
702	if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)) ||
703	    flags == MCL_ONFAULT)
704		return -EINVAL;
705
706	if (!can_do_mlock())
707		return -EPERM;
708
709	lock_limit = rlimit(RLIMIT_MEMLOCK);
710	lock_limit >>= PAGE_SHIFT;
711
712	if (mmap_write_lock_killable(current->mm))
713		return -EINTR;
714
715	ret = -ENOMEM;
716	if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
717	    capable(CAP_IPC_LOCK))
718		ret = apply_mlockall_flags(flags);
719	mmap_write_unlock(current->mm);
720	if (!ret && (flags & MCL_CURRENT))
721		mm_populate(0, TASK_SIZE);
722
723	return ret;
724}
725
726SYSCALL_DEFINE0(munlockall)
727{
728	int ret;
729
730	if (mmap_write_lock_killable(current->mm))
731		return -EINTR;
732	ret = apply_mlockall_flags(0);
733	mmap_write_unlock(current->mm);
734	return ret;
735}
736
737/*
738 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
739 * shm segments) get accounted against the user_struct instead.
740 */
741static DEFINE_SPINLOCK(shmlock_user_lock);
742
743int user_shm_lock(size_t size, struct ucounts *ucounts)
744{
745	unsigned long lock_limit, locked;
746	long memlock;
747	int allowed = 0;
748
749	locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
750	lock_limit = rlimit(RLIMIT_MEMLOCK);
751	if (lock_limit != RLIM_INFINITY)
752		lock_limit >>= PAGE_SHIFT;
 
753	spin_lock(&shmlock_user_lock);
754	memlock = inc_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
755
756	if ((memlock == LONG_MAX || memlock > lock_limit) && !capable(CAP_IPC_LOCK)) {
757		dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
758		goto out;
759	}
760	if (!get_ucounts(ucounts)) {
761		dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
762		allowed = 0;
763		goto out;
764	}
765	allowed = 1;
766out:
767	spin_unlock(&shmlock_user_lock);
768	return allowed;
769}
770
771void user_shm_unlock(size_t size, struct ucounts *ucounts)
772{
773	spin_lock(&shmlock_user_lock);
774	dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, (size + PAGE_SIZE - 1) >> PAGE_SHIFT);
775	spin_unlock(&shmlock_user_lock);
776	put_ucounts(ucounts);
777}
v5.9
  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/mempolicy.h>
 18#include <linux/syscalls.h>
 19#include <linux/sched.h>
 20#include <linux/export.h>
 21#include <linux/rmap.h>
 22#include <linux/mmzone.h>
 23#include <linux/hugetlb.h>
 24#include <linux/memcontrol.h>
 25#include <linux/mm_inline.h>
 
 26
 27#include "internal.h"
 28
 
 
 
 
 
 
 
 
 
 29bool can_do_mlock(void)
 30{
 31	if (rlimit(RLIMIT_MEMLOCK) != 0)
 32		return true;
 33	if (capable(CAP_IPC_LOCK))
 34		return true;
 35	return false;
 36}
 37EXPORT_SYMBOL(can_do_mlock);
 38
 39/*
 40 * Mlocked pages are marked with PageMlocked() flag for efficient testing
 41 * in vmscan and, possibly, the fault path; and to support semi-accurate
 42 * statistics.
 43 *
 44 * An mlocked page [PageMlocked(page)] is unevictable.  As such, it will
 45 * be placed on the LRU "unevictable" list, rather than the [in]active lists.
 46 * The unevictable list is an LRU sibling list to the [in]active lists.
 47 * PageUnevictable is set to indicate the unevictable state.
 48 *
 49 * When lazy mlocking via vmscan, it is important to ensure that the
 50 * vma's VM_LOCKED status is not concurrently being modified, otherwise we
 51 * may have mlocked a page that is being munlocked. So lazy mlock must take
 52 * the mmap_lock for read, and verify that the vma really is locked
 53 * (see mm/rmap.c).
 54 */
 55
 56/*
 57 *  LRU accounting for clear_page_mlock()
 58 */
 59void clear_page_mlock(struct page *page)
 60{
 61	int nr_pages;
 
 
 62
 63	if (!TestClearPageMlocked(page))
 64		return;
 65
 66	nr_pages = thp_nr_pages(page);
 67	mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
 68	count_vm_events(UNEVICTABLE_PGCLEARED, nr_pages);
 69	/*
 70	 * The previous TestClearPageMlocked() corresponds to the smp_mb()
 71	 * in __pagevec_lru_add_fn().
 72	 *
 73	 * See __pagevec_lru_add_fn for more explanation.
 74	 */
 75	if (!isolate_lru_page(page)) {
 76		putback_lru_page(page);
 77	} else {
 78		/*
 79		 * We lost the race. the page already moved to evictable list.
 
 
 
 80		 */
 81		if (PageUnevictable(page))
 82			count_vm_events(UNEVICTABLE_PGSTRANDED, nr_pages);
 
 
 
 
 
 
 83	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 84}
 85
 86/*
 87 * Mark page as mlocked if not already.
 88 * If page on LRU, isolate and putback to move to unevictable list.
 89 */
 90void mlock_vma_page(struct page *page)
 91{
 92	/* Serialize with page migration */
 93	BUG_ON(!PageLocked(page));
 94
 95	VM_BUG_ON_PAGE(PageTail(page), page);
 96	VM_BUG_ON_PAGE(PageCompound(page) && PageDoubleMap(page), page);
 97
 98	if (!TestSetPageMlocked(page)) {
 99		int nr_pages = thp_nr_pages(page);
 
100
101		mod_zone_page_state(page_zone(page), NR_MLOCK, nr_pages);
102		count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
103		if (!isolate_lru_page(page))
104			putback_lru_page(page);
105	}
 
 
106}
107
108/*
109 * Isolate a page from LRU with optional get_page() pin.
110 * Assumes lru_lock already held and page already pinned.
111 */
112static bool __munlock_isolate_lru_page(struct page *page, bool getpage)
113{
114	if (PageLRU(page)) {
115		struct lruvec *lruvec;
116
117		lruvec = mem_cgroup_page_lruvec(page, page_pgdat(page));
118		if (getpage)
119			get_page(page);
120		ClearPageLRU(page);
121		del_page_from_lru_list(page, lruvec, page_lru(page));
122		return true;
 
 
 
 
 
 
123	}
 
124
125	return false;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
126}
127
128/*
129 * Finish munlock after successful page isolation
130 *
131 * Page must be locked. This is a wrapper for try_to_munlock()
132 * and putback_lru_page() with munlock accounting.
133 */
134static void __munlock_isolated_page(struct page *page)
 
 
135{
136	/*
137	 * Optimization: if the page was mapped just once, that's our mapping
138	 * and we don't need to check all the other vmas.
139	 */
140	if (page_mapcount(page) > 1)
141		try_to_munlock(page);
142
143	/* Did try_to_unlock() succeed or punt? */
144	if (!PageMlocked(page))
145		count_vm_events(UNEVICTABLE_PGMUNLOCKED, thp_nr_pages(page));
146
147	putback_lru_page(page);
148}
149
150/*
151 * Accounting for page isolation fail during munlock
152 *
153 * Performs accounting when page isolation fails in munlock. There is nothing
154 * else to do because it means some other task has already removed the page
155 * from the LRU. putback_lru_page() will take care of removing the page from
156 * the unevictable list, if necessary. vmscan [page_referenced()] will move
157 * the page back to the unevictable list if some other vma has it mlocked.
158 */
159static void __munlock_isolation_failed(struct page *page)
160{
161	int nr_pages = thp_nr_pages(page);
162
163	if (PageUnevictable(page))
164		__count_vm_events(UNEVICTABLE_PGSTRANDED, nr_pages);
165	else
166		__count_vm_events(UNEVICTABLE_PGMUNLOCKED, nr_pages);
167}
168
169/**
170 * munlock_vma_page - munlock a vma page
171 * @page: page to be unlocked, either a normal page or THP page head
172 *
173 * returns the size of the page as a page mask (0 for normal page,
174 *         HPAGE_PMD_NR - 1 for THP head page)
175 *
176 * called from munlock()/munmap() path with page supposedly on the LRU.
177 * When we munlock a page, because the vma where we found the page is being
178 * munlock()ed or munmap()ed, we want to check whether other vmas hold the
179 * page locked so that we can leave it on the unevictable lru list and not
180 * bother vmscan with it.  However, to walk the page's rmap list in
181 * try_to_munlock() we must isolate the page from the LRU.  If some other
182 * task has removed the page from the LRU, we won't be able to do that.
183 * So we clear the PageMlocked as we might not get another chance.  If we
184 * can't isolate the page, we leave it for putback_lru_page() and vmscan
185 * [page_referenced()/try_to_unmap()] to deal with.
186 */
187unsigned int munlock_vma_page(struct page *page)
188{
189	int nr_pages;
190	pg_data_t *pgdat = page_pgdat(page);
 
 
191
192	/* For try_to_munlock() and to serialize with page migration */
193	BUG_ON(!PageLocked(page));
 
 
 
194
195	VM_BUG_ON_PAGE(PageTail(page), page);
 
 
 
 
 
 
196
197	/*
198	 * Serialize with any parallel __split_huge_page_refcount() which
199	 * might otherwise copy PageMlocked to part of the tail pages before
200	 * we clear it in the head page. It also stabilizes thp_nr_pages().
201	 */
202	spin_lock_irq(&pgdat->lru_lock);
203
204	if (!TestClearPageMlocked(page)) {
205		/* Potentially, PTE-mapped THP: do not skip the rest PTEs */
206		nr_pages = 1;
207		goto unlock_out;
208	}
209
210	nr_pages = thp_nr_pages(page);
211	__mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
 
 
 
 
212
213	if (__munlock_isolate_lru_page(page, true)) {
214		spin_unlock_irq(&pgdat->lru_lock);
215		__munlock_isolated_page(page);
216		goto out;
217	}
218	__munlock_isolation_failed(page);
219
220unlock_out:
221	spin_unlock_irq(&pgdat->lru_lock);
222
223out:
224	return nr_pages - 1;
225}
226
227/*
228 * convert get_user_pages() return value to posix mlock() error
229 */
230static int __mlock_posix_error_return(long retval)
231{
232	if (retval == -EFAULT)
233		retval = -ENOMEM;
234	else if (retval == -ENOMEM)
235		retval = -EAGAIN;
236	return retval;
237}
238
239/*
240 * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec()
241 *
242 * The fast path is available only for evictable pages with single mapping.
243 * Then we can bypass the per-cpu pvec and get better performance.
244 * when mapcount > 1 we need try_to_munlock() which can fail.
245 * when !page_evictable(), we need the full redo logic of putback_lru_page to
246 * avoid leaving evictable page in unevictable list.
247 *
248 * In case of success, @page is added to @pvec and @pgrescued is incremented
249 * in case that the page was previously unevictable. @page is also unlocked.
250 */
251static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec,
252		int *pgrescued)
253{
254	VM_BUG_ON_PAGE(PageLRU(page), page);
255	VM_BUG_ON_PAGE(!PageLocked(page), page);
 
 
 
 
 
256
257	if (page_mapcount(page) <= 1 && page_evictable(page)) {
258		pagevec_add(pvec, page);
259		if (TestClearPageUnevictable(page))
260			(*pgrescued)++;
261		unlock_page(page);
262		return true;
263	}
264
265	return false;
 
 
 
 
266}
267
268/*
269 * Putback multiple evictable pages to the LRU
270 *
271 * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of
272 * the pages might have meanwhile become unevictable but that is OK.
273 */
274static void __putback_lru_fast(struct pagevec *pvec, int pgrescued)
275{
276	count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec));
277	/*
278	 *__pagevec_lru_add() calls release_pages() so we don't call
279	 * put_page() explicitly
280	 */
281	__pagevec_lru_add(pvec);
282	count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued);
 
 
 
 
 
 
 
283}
284
285/*
286 * Munlock a batch of pages from the same zone
287 *
288 * The work is split to two main phases. First phase clears the Mlocked flag
289 * and attempts to isolate the pages, all under a single zone lru lock.
290 * The second phase finishes the munlock only for pages where isolation
291 * succeeded.
292 *
293 * Note that the pagevec may be modified during the process.
294 */
295static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone)
296{
297	int i;
298	int nr = pagevec_count(pvec);
299	int delta_munlocked = -nr;
300	struct pagevec pvec_putback;
301	int pgrescued = 0;
302
303	pagevec_init(&pvec_putback);
304
305	/* Phase 1: page isolation */
306	spin_lock_irq(&zone->zone_pgdat->lru_lock);
307	for (i = 0; i < nr; i++) {
308		struct page *page = pvec->pages[i];
309
310		if (TestClearPageMlocked(page)) {
311			/*
312			 * We already have pin from follow_page_mask()
313			 * so we can spare the get_page() here.
314			 */
315			if (__munlock_isolate_lru_page(page, false))
316				continue;
317			else
318				__munlock_isolation_failed(page);
319		} else {
320			delta_munlocked++;
321		}
322
323		/*
324		 * We won't be munlocking this page in the next phase
325		 * but we still need to release the follow_page_mask()
326		 * pin. We cannot do it under lru_lock however. If it's
327		 * the last pin, __page_cache_release() would deadlock.
328		 */
329		pagevec_add(&pvec_putback, pvec->pages[i]);
330		pvec->pages[i] = NULL;
331	}
332	__mod_zone_page_state(zone, NR_MLOCK, delta_munlocked);
333	spin_unlock_irq(&zone->zone_pgdat->lru_lock);
334
335	/* Now we can release pins of pages that we are not munlocking */
336	pagevec_release(&pvec_putback);
337
338	/* Phase 2: page munlock */
339	for (i = 0; i < nr; i++) {
340		struct page *page = pvec->pages[i];
341
342		if (page) {
343			lock_page(page);
344			if (!__putback_lru_fast_prepare(page, &pvec_putback,
345					&pgrescued)) {
346				/*
347				 * Slow path. We don't want to lose the last
348				 * pin before unlock_page()
349				 */
350				get_page(page); /* for putback_lru_page() */
351				__munlock_isolated_page(page);
352				unlock_page(page);
353				put_page(page); /* from follow_page_mask() */
354			}
355		}
356	}
357
 
 
358	/*
359	 * Phase 3: page putback for pages that qualified for the fast path
360	 * This will also call put_page() to return pin from follow_page_mask()
361	 */
362	if (pagevec_count(&pvec_putback))
363		__putback_lru_fast(&pvec_putback, pgrescued);
 
 
 
 
364}
365
366/*
367 * Fill up pagevec for __munlock_pagevec using pte walk
368 *
369 * The function expects that the struct page corresponding to @start address is
370 * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone.
371 *
372 * The rest of @pvec is filled by subsequent pages within the same pmd and same
373 * zone, as long as the pte's are present and vm_normal_page() succeeds. These
374 * pages also get pinned.
375 *
376 * Returns the address of the next page that should be scanned. This equals
377 * @start + PAGE_SIZE when no page could be added by the pte walk.
378 */
379static unsigned long __munlock_pagevec_fill(struct pagevec *pvec,
380			struct vm_area_struct *vma, struct zone *zone,
381			unsigned long start, unsigned long end)
382{
383	pte_t *pte;
384	spinlock_t *ptl;
 
 
385
386	/*
387	 * Initialize pte walk starting at the already pinned page where we
388	 * are sure that there is a pte, as it was pinned under the same
389	 * mmap_lock write op.
390	 */
391	pte = get_locked_pte(vma->vm_mm, start,	&ptl);
392	/* Make sure we do not cross the page table boundary */
393	end = pgd_addr_end(start, end);
394	end = p4d_addr_end(start, end);
395	end = pud_addr_end(start, end);
396	end = pmd_addr_end(start, end);
397
398	/* The page next to the pinned page is the first we will try to get */
399	start += PAGE_SIZE;
400	while (start < end) {
401		struct page *page = NULL;
402		pte++;
403		if (pte_present(*pte))
404			page = vm_normal_page(vma, start, *pte);
405		/*
406		 * Break if page could not be obtained or the page's node+zone does not
407		 * match
408		 */
409		if (!page || page_zone(page) != zone)
410			break;
411
412		/*
413		 * Do not use pagevec for PTE-mapped THP,
414		 * munlock_vma_pages_range() will handle them.
415		 */
 
 
 
416		if (PageTransCompound(page))
417			break;
418
419		get_page(page);
420		/*
421		 * Increase the address that will be returned *before* the
422		 * eventual break due to pvec becoming full by adding the page
423		 */
424		start += PAGE_SIZE;
425		if (pagevec_add(pvec, page) == 0)
426			break;
427	}
428	pte_unmap_unlock(pte, ptl);
429	return start;
 
 
 
430}
431
432/*
433 * munlock_vma_pages_range() - munlock all pages in the vma range.'
434 * @vma - vma containing range to be munlock()ed.
 
435 * @start - start address in @vma of the range
436 * @end - end of range in @vma.
 
437 *
438 *  For mremap(), munmap() and exit().
439 *
440 * Called with @vma VM_LOCKED.
441 *
442 * Returns with VM_LOCKED cleared.  Callers must be prepared to
443 * deal with this.
444 *
445 * We don't save and restore VM_LOCKED here because pages are
446 * still on lru.  In unmap path, pages might be scanned by reclaim
447 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
448 * free them.  This will result in freeing mlocked pages.
449 */
450void munlock_vma_pages_range(struct vm_area_struct *vma,
451			     unsigned long start, unsigned long end)
452{
453	vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
454
455	while (start < end) {
456		struct page *page;
457		unsigned int page_mask = 0;
458		unsigned long page_increm;
459		struct pagevec pvec;
460		struct zone *zone;
461
462		pagevec_init(&pvec);
463		/*
464		 * Although FOLL_DUMP is intended for get_dump_page(),
465		 * it just so happens that its special treatment of the
466		 * ZERO_PAGE (returning an error instead of doing get_page)
467		 * suits munlock very well (and if somehow an abnormal page
468		 * has sneaked into the range, we won't oops here: great).
469		 */
470		page = follow_page(vma, start, FOLL_GET | FOLL_DUMP);
471
472		if (page && !IS_ERR(page)) {
473			if (PageTransTail(page)) {
474				VM_BUG_ON_PAGE(PageMlocked(page), page);
475				put_page(page); /* follow_page_mask() */
476			} else if (PageTransHuge(page)) {
477				lock_page(page);
478				/*
479				 * Any THP page found by follow_page_mask() may
480				 * have gotten split before reaching
481				 * munlock_vma_page(), so we need to compute
482				 * the page_mask here instead.
483				 */
484				page_mask = munlock_vma_page(page);
485				unlock_page(page);
486				put_page(page); /* follow_page_mask() */
487			} else {
488				/*
489				 * Non-huge pages are handled in batches via
490				 * pagevec. The pin from follow_page_mask()
491				 * prevents them from collapsing by THP.
492				 */
493				pagevec_add(&pvec, page);
494				zone = page_zone(page);
495
496				/*
497				 * Try to fill the rest of pagevec using fast
498				 * pte walk. This will also update start to
499				 * the next page to process. Then munlock the
500				 * pagevec.
501				 */
502				start = __munlock_pagevec_fill(&pvec, vma,
503						zone, start, end);
504				__munlock_pagevec(&pvec, zone);
505				goto next;
506			}
507		}
508		page_increm = 1 + page_mask;
509		start += page_increm * PAGE_SIZE;
510next:
511		cond_resched();
512	}
513}
514
515/*
516 * mlock_fixup  - handle mlock[all]/munlock[all] requests.
517 *
518 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
519 * munlock is a no-op.  However, for some special vmas, we go ahead and
520 * populate the ptes.
521 *
522 * For vmas that pass the filters, merge/split as appropriate.
523 */
524static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
525	unsigned long start, unsigned long end, vm_flags_t newflags)
526{
527	struct mm_struct *mm = vma->vm_mm;
528	pgoff_t pgoff;
529	int nr_pages;
530	int ret = 0;
531	int lock = !!(newflags & VM_LOCKED);
532	vm_flags_t old_flags = vma->vm_flags;
533
534	if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
535	    is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm) ||
536	    vma_is_dax(vma))
537		/* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
538		goto out;
539
540	pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
541	*prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
542			  vma->vm_file, pgoff, vma_policy(vma),
543			  vma->vm_userfaultfd_ctx);
544	if (*prev) {
545		vma = *prev;
546		goto success;
547	}
548
549	if (start != vma->vm_start) {
550		ret = split_vma(mm, vma, start, 1);
551		if (ret)
552			goto out;
553	}
554
555	if (end != vma->vm_end) {
556		ret = split_vma(mm, vma, end, 0);
557		if (ret)
558			goto out;
559	}
560
561success:
562	/*
563	 * Keep track of amount of locked VM.
564	 */
565	nr_pages = (end - start) >> PAGE_SHIFT;
566	if (!lock)
567		nr_pages = -nr_pages;
568	else if (old_flags & VM_LOCKED)
569		nr_pages = 0;
570	mm->locked_vm += nr_pages;
571
572	/*
573	 * vm_flags is protected by the mmap_lock held in write mode.
574	 * It's okay if try_to_unmap_one unmaps a page just after we
575	 * set VM_LOCKED, populate_vma_page_range will bring it back.
576	 */
577
578	if (lock)
 
579		vma->vm_flags = newflags;
580	else
581		munlock_vma_pages_range(vma, start, end);
582
583out:
584	*prev = vma;
585	return ret;
586}
587
588static int apply_vma_lock_flags(unsigned long start, size_t len,
589				vm_flags_t flags)
590{
591	unsigned long nstart, end, tmp;
592	struct vm_area_struct * vma, * prev;
593	int error;
 
594
595	VM_BUG_ON(offset_in_page(start));
596	VM_BUG_ON(len != PAGE_ALIGN(len));
597	end = start + len;
598	if (end < start)
599		return -EINVAL;
600	if (end == start)
601		return 0;
602	vma = find_vma(current->mm, start);
603	if (!vma || vma->vm_start > start)
604		return -ENOMEM;
605
606	prev = vma->vm_prev;
607	if (start > vma->vm_start)
608		prev = vma;
 
 
609
610	for (nstart = start ; ; ) {
611		vm_flags_t newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
612
613		newflags |= flags;
614
615		/* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
616		tmp = vma->vm_end;
617		if (tmp > end)
618			tmp = end;
619		error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
620		if (error)
621			break;
622		nstart = tmp;
623		if (nstart < prev->vm_end)
624			nstart = prev->vm_end;
625		if (nstart >= end)
626			break;
627
628		vma = prev->vm_next;
629		if (!vma || vma->vm_start != nstart) {
630			error = -ENOMEM;
631			break;
632		}
633	}
634	return error;
635}
636
637/*
638 * Go through vma areas and sum size of mlocked
639 * vma pages, as return value.
640 * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT)
641 * is also counted.
642 * Return value: previously mlocked page counts
643 */
644static unsigned long count_mm_mlocked_page_nr(struct mm_struct *mm,
645		unsigned long start, size_t len)
646{
647	struct vm_area_struct *vma;
648	unsigned long count = 0;
 
 
649
650	if (mm == NULL)
651		mm = current->mm;
652
653	vma = find_vma(mm, start);
654	if (vma == NULL)
655		vma = mm->mmap;
656
657	for (; vma ; vma = vma->vm_next) {
658		if (start >= vma->vm_end)
659			continue;
660		if (start + len <=  vma->vm_start)
661			break;
662		if (vma->vm_flags & VM_LOCKED) {
663			if (start > vma->vm_start)
664				count -= (start - vma->vm_start);
665			if (start + len < vma->vm_end) {
666				count += start + len - vma->vm_start;
667				break;
668			}
669			count += vma->vm_end - vma->vm_start;
670		}
671	}
672
673	return count >> PAGE_SHIFT;
674}
675
 
 
 
 
 
 
 
 
 
 
 
 
676static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
677{
678	unsigned long locked;
679	unsigned long lock_limit;
680	int error = -ENOMEM;
681
682	start = untagged_addr(start);
683
684	if (!can_do_mlock())
685		return -EPERM;
686
687	len = PAGE_ALIGN(len + (offset_in_page(start)));
688	start &= PAGE_MASK;
689
690	lock_limit = rlimit(RLIMIT_MEMLOCK);
691	lock_limit >>= PAGE_SHIFT;
692	locked = len >> PAGE_SHIFT;
693
694	if (mmap_write_lock_killable(current->mm))
695		return -EINTR;
696
697	locked += current->mm->locked_vm;
698	if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) {
699		/*
700		 * It is possible that the regions requested intersect with
701		 * previously mlocked areas, that part area in "mm->locked_vm"
702		 * should not be counted to new mlock increment count. So check
703		 * and adjust locked count if necessary.
704		 */
705		locked -= count_mm_mlocked_page_nr(current->mm,
706				start, len);
707	}
708
709	/* check against resource limits */
710	if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
711		error = apply_vma_lock_flags(start, len, flags);
712
713	mmap_write_unlock(current->mm);
714	if (error)
715		return error;
716
717	error = __mm_populate(start, len, 0);
718	if (error)
719		return __mlock_posix_error_return(error);
720	return 0;
721}
722
723SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
724{
725	return do_mlock(start, len, VM_LOCKED);
726}
727
728SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
729{
730	vm_flags_t vm_flags = VM_LOCKED;
731
732	if (flags & ~MLOCK_ONFAULT)
733		return -EINVAL;
734
735	if (flags & MLOCK_ONFAULT)
736		vm_flags |= VM_LOCKONFAULT;
737
738	return do_mlock(start, len, vm_flags);
739}
740
741SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
742{
743	int ret;
744
745	start = untagged_addr(start);
746
747	len = PAGE_ALIGN(len + (offset_in_page(start)));
748	start &= PAGE_MASK;
749
750	if (mmap_write_lock_killable(current->mm))
751		return -EINTR;
752	ret = apply_vma_lock_flags(start, len, 0);
753	mmap_write_unlock(current->mm);
754
755	return ret;
756}
757
758/*
759 * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
760 * and translate into the appropriate modifications to mm->def_flags and/or the
761 * flags for all current VMAs.
762 *
763 * There are a couple of subtleties with this.  If mlockall() is called multiple
764 * times with different flags, the values do not necessarily stack.  If mlockall
765 * is called once including the MCL_FUTURE flag and then a second time without
766 * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
767 */
768static int apply_mlockall_flags(int flags)
769{
770	struct vm_area_struct * vma, * prev = NULL;
 
771	vm_flags_t to_add = 0;
772
773	current->mm->def_flags &= VM_LOCKED_CLEAR_MASK;
774	if (flags & MCL_FUTURE) {
775		current->mm->def_flags |= VM_LOCKED;
776
777		if (flags & MCL_ONFAULT)
778			current->mm->def_flags |= VM_LOCKONFAULT;
779
780		if (!(flags & MCL_CURRENT))
781			goto out;
782	}
783
784	if (flags & MCL_CURRENT) {
785		to_add |= VM_LOCKED;
786		if (flags & MCL_ONFAULT)
787			to_add |= VM_LOCKONFAULT;
788	}
789
790	for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
791		vm_flags_t newflags;
792
793		newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
794		newflags |= to_add;
795
796		/* Ignore errors */
797		mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
 
798		cond_resched();
799	}
800out:
801	return 0;
802}
803
804SYSCALL_DEFINE1(mlockall, int, flags)
805{
806	unsigned long lock_limit;
807	int ret;
808
809	if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)) ||
810	    flags == MCL_ONFAULT)
811		return -EINVAL;
812
813	if (!can_do_mlock())
814		return -EPERM;
815
816	lock_limit = rlimit(RLIMIT_MEMLOCK);
817	lock_limit >>= PAGE_SHIFT;
818
819	if (mmap_write_lock_killable(current->mm))
820		return -EINTR;
821
822	ret = -ENOMEM;
823	if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
824	    capable(CAP_IPC_LOCK))
825		ret = apply_mlockall_flags(flags);
826	mmap_write_unlock(current->mm);
827	if (!ret && (flags & MCL_CURRENT))
828		mm_populate(0, TASK_SIZE);
829
830	return ret;
831}
832
833SYSCALL_DEFINE0(munlockall)
834{
835	int ret;
836
837	if (mmap_write_lock_killable(current->mm))
838		return -EINTR;
839	ret = apply_mlockall_flags(0);
840	mmap_write_unlock(current->mm);
841	return ret;
842}
843
844/*
845 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
846 * shm segments) get accounted against the user_struct instead.
847 */
848static DEFINE_SPINLOCK(shmlock_user_lock);
849
850int user_shm_lock(size_t size, struct user_struct *user)
851{
852	unsigned long lock_limit, locked;
 
853	int allowed = 0;
854
855	locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
856	lock_limit = rlimit(RLIMIT_MEMLOCK);
857	if (lock_limit == RLIM_INFINITY)
858		allowed = 1;
859	lock_limit >>= PAGE_SHIFT;
860	spin_lock(&shmlock_user_lock);
861	if (!allowed &&
862	    locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
 
 
863		goto out;
864	get_uid(user);
865	user->locked_shm += locked;
 
 
 
 
866	allowed = 1;
867out:
868	spin_unlock(&shmlock_user_lock);
869	return allowed;
870}
871
872void user_shm_unlock(size_t size, struct user_struct *user)
873{
874	spin_lock(&shmlock_user_lock);
875	user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
876	spin_unlock(&shmlock_user_lock);
877	free_uid(user);
878}