Linux Audio

Check our new training course

Loading...
v5.14.15
  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#include <linux/secretmem.h>
 27
 28#include "internal.h"
 29
 
 
 
 
 
 
 
 
 
 30bool can_do_mlock(void)
 31{
 32	if (rlimit(RLIMIT_MEMLOCK) != 0)
 33		return true;
 34	if (capable(CAP_IPC_LOCK))
 35		return true;
 36	return false;
 37}
 38EXPORT_SYMBOL(can_do_mlock);
 39
 40/*
 41 * Mlocked pages are marked with PageMlocked() flag for efficient testing
 42 * in vmscan and, possibly, the fault path; and to support semi-accurate
 43 * statistics.
 44 *
 45 * An mlocked page [PageMlocked(page)] is unevictable.  As such, it will
 46 * be placed on the LRU "unevictable" list, rather than the [in]active lists.
 47 * The unevictable list is an LRU sibling list to the [in]active lists.
 48 * PageUnevictable is set to indicate the unevictable state.
 49 *
 50 * When lazy mlocking via vmscan, it is important to ensure that the
 51 * vma's VM_LOCKED status is not concurrently being modified, otherwise we
 52 * may have mlocked a page that is being munlocked. So lazy mlock must take
 53 * the mmap_lock for read, and verify that the vma really is locked
 54 * (see mm/rmap.c).
 55 */
 56
 57/*
 58 *  LRU accounting for clear_page_mlock()
 59 */
 60void clear_page_mlock(struct page *page)
 61{
 62	int nr_pages;
 
 
 63
 64	if (!TestClearPageMlocked(page))
 65		return;
 66
 67	nr_pages = thp_nr_pages(page);
 68	mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
 69	count_vm_events(UNEVICTABLE_PGCLEARED, nr_pages);
 70	/*
 71	 * The previous TestClearPageMlocked() corresponds to the smp_mb()
 72	 * in __pagevec_lru_add_fn().
 73	 *
 74	 * See __pagevec_lru_add_fn for more explanation.
 75	 */
 76	if (!isolate_lru_page(page)) {
 77		putback_lru_page(page);
 78	} else {
 79		/*
 80		 * We lost the race. the page already moved to evictable list.
 
 
 81		 */
 82		if (PageUnevictable(page))
 83			count_vm_events(UNEVICTABLE_PGSTRANDED, nr_pages);
 
 
 
 
 
 
 
 84	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 85}
 86
 87/*
 88 * Mark page as mlocked if not already.
 89 * If page on LRU, isolate and putback to move to unevictable list.
 90 */
 91void mlock_vma_page(struct page *page)
 92{
 93	/* Serialize with page migration */
 94	BUG_ON(!PageLocked(page));
 95
 96	VM_BUG_ON_PAGE(PageTail(page), page);
 97	VM_BUG_ON_PAGE(PageCompound(page) && PageDoubleMap(page), page);
 98
 99	if (!TestSetPageMlocked(page)) {
100		int nr_pages = thp_nr_pages(page);
 
101
102		mod_zone_page_state(page_zone(page), NR_MLOCK, nr_pages);
103		count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
104		if (!isolate_lru_page(page))
105			putback_lru_page(page);
106	}
 
 
107}
108
109/*
110 * Finish munlock after successful page isolation
111 *
112 * Page must be locked. This is a wrapper for page_mlock()
113 * and putback_lru_page() with munlock accounting.
114 */
115static void __munlock_isolated_page(struct page *page)
116{
117	/*
118	 * Optimization: if the page was mapped just once, that's our mapping
119	 * and we don't need to check all the other vmas.
120	 */
121	if (page_mapcount(page) > 1)
122		page_mlock(page);
123
124	/* Did try_to_unlock() succeed or punt? */
125	if (!PageMlocked(page))
126		count_vm_events(UNEVICTABLE_PGMUNLOCKED, thp_nr_pages(page));
 
 
 
 
 
 
 
 
 
 
 
127
128	putback_lru_page(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
129}
130
131/*
132 * Accounting for page isolation fail during munlock
133 *
134 * Performs accounting when page isolation fails in munlock. There is nothing
135 * else to do because it means some other task has already removed the page
136 * from the LRU. putback_lru_page() will take care of removing the page from
137 * the unevictable list, if necessary. vmscan [page_referenced()] will move
138 * the page back to the unevictable list if some other vma has it mlocked.
139 */
140static void __munlock_isolation_failed(struct page *page)
 
 
141{
142	int nr_pages = thp_nr_pages(page);
 
143
144	if (PageUnevictable(page))
145		__count_vm_events(UNEVICTABLE_PGSTRANDED, nr_pages);
146	else
147		__count_vm_events(UNEVICTABLE_PGMUNLOCKED, nr_pages);
148}
149
150/**
151 * munlock_vma_page - munlock a vma page
152 * @page: page to be unlocked, either a normal page or THP page head
153 *
154 * returns the size of the page as a page mask (0 for normal page,
155 *         HPAGE_PMD_NR - 1 for THP head page)
156 *
157 * called from munlock()/munmap() path with page supposedly on the LRU.
158 * When we munlock a page, because the vma where we found the page is being
159 * munlock()ed or munmap()ed, we want to check whether other vmas hold the
160 * page locked so that we can leave it on the unevictable lru list and not
161 * bother vmscan with it.  However, to walk the page's rmap list in
162 * page_mlock() we must isolate the page from the LRU.  If some other
163 * task has removed the page from the LRU, we won't be able to do that.
164 * So we clear the PageMlocked as we might not get another chance.  If we
165 * can't isolate the page, we leave it for putback_lru_page() and vmscan
166 * [page_referenced()/try_to_unmap()] to deal with.
167 */
168unsigned int munlock_vma_page(struct page *page)
169{
170	int nr_pages;
171
172	/* For page_mlock() and to serialize with page migration */
173	BUG_ON(!PageLocked(page));
174	VM_BUG_ON_PAGE(PageTail(page), page);
175
176	if (!TestClearPageMlocked(page)) {
177		/* Potentially, PTE-mapped THP: do not skip the rest PTEs */
178		return 0;
 
 
 
 
 
 
 
 
 
179	}
180
181	nr_pages = thp_nr_pages(page);
182	mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
 
 
 
 
 
 
183
184	if (!isolate_lru_page(page))
185		__munlock_isolated_page(page);
186	else
187		__munlock_isolation_failed(page);
 
 
 
 
 
 
188
189	return nr_pages - 1;
 
 
 
190}
191
192/*
193 * convert get_user_pages() return value to posix mlock() error
194 */
195static int __mlock_posix_error_return(long retval)
196{
197	if (retval == -EFAULT)
198		retval = -ENOMEM;
199	else if (retval == -ENOMEM)
200		retval = -EAGAIN;
201	return retval;
202}
203
204/*
205 * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec()
206 *
207 * The fast path is available only for evictable pages with single mapping.
208 * Then we can bypass the per-cpu pvec and get better performance.
209 * when mapcount > 1 we need page_mlock() which can fail.
210 * when !page_evictable(), we need the full redo logic of putback_lru_page to
211 * avoid leaving evictable page in unevictable list.
212 *
213 * In case of success, @page is added to @pvec and @pgrescued is incremented
214 * in case that the page was previously unevictable. @page is also unlocked.
215 */
216static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec,
217		int *pgrescued)
218{
219	VM_BUG_ON_PAGE(PageLRU(page), page);
220	VM_BUG_ON_PAGE(!PageLocked(page), page);
221
222	if (page_mapcount(page) <= 1 && page_evictable(page)) {
223		pagevec_add(pvec, page);
224		if (TestClearPageUnevictable(page))
225			(*pgrescued)++;
226		unlock_page(page);
227		return true;
 
 
228	}
229
230	return false;
 
 
 
 
231}
232
233/*
234 * Putback multiple evictable pages to the LRU
235 *
236 * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of
237 * the pages might have meanwhile become unevictable but that is OK.
238 */
239static void __putback_lru_fast(struct pagevec *pvec, int pgrescued)
240{
241	count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec));
242	/*
243	 *__pagevec_lru_add() calls release_pages() so we don't call
244	 * put_page() explicitly
245	 */
246	__pagevec_lru_add(pvec);
247	count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued);
 
 
 
 
 
 
 
 
248}
249
250/*
251 * Munlock a batch of pages from the same zone
252 *
253 * The work is split to two main phases. First phase clears the Mlocked flag
254 * and attempts to isolate the pages, all under a single zone lru lock.
255 * The second phase finishes the munlock only for pages where isolation
256 * succeeded.
257 *
258 * Note that the pagevec may be modified during the process.
259 */
260static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone)
261{
262	int i;
263	int nr = pagevec_count(pvec);
264	int delta_munlocked = -nr;
265	struct pagevec pvec_putback;
266	struct lruvec *lruvec = NULL;
267	int pgrescued = 0;
 
 
 
 
 
 
 
 
268
269	pagevec_init(&pvec_putback);
 
 
 
 
 
270
271	/* Phase 1: page isolation */
272	for (i = 0; i < nr; i++) {
273		struct page *page = pvec->pages[i];
274
275		if (TestClearPageMlocked(page)) {
276			/*
277			 * We already have pin from follow_page_mask()
278			 * so we can spare the get_page() here.
279			 */
280			if (TestClearPageLRU(page)) {
281				lruvec = relock_page_lruvec_irq(page, lruvec);
282				del_page_from_lru_list(page, lruvec);
283				continue;
284			} else
285				__munlock_isolation_failed(page);
286		} else {
287			delta_munlocked++;
288		}
289
290		/*
291		 * We won't be munlocking this page in the next phase
292		 * but we still need to release the follow_page_mask()
293		 * pin. We cannot do it under lru_lock however. If it's
294		 * the last pin, __page_cache_release() would deadlock.
295		 */
296		pagevec_add(&pvec_putback, pvec->pages[i]);
297		pvec->pages[i] = NULL;
298	}
299	if (lruvec) {
300		__mod_zone_page_state(zone, NR_MLOCK, delta_munlocked);
301		unlock_page_lruvec_irq(lruvec);
302	} else if (delta_munlocked) {
303		mod_zone_page_state(zone, NR_MLOCK, delta_munlocked);
304	}
305
306	/* Now we can release pins of pages that we are not munlocking */
307	pagevec_release(&pvec_putback);
308
309	/* Phase 2: page munlock */
310	for (i = 0; i < nr; i++) {
311		struct page *page = pvec->pages[i];
312
313		if (page) {
314			lock_page(page);
315			if (!__putback_lru_fast_prepare(page, &pvec_putback,
316					&pgrescued)) {
317				/*
318				 * Slow path. We don't want to lose the last
319				 * pin before unlock_page()
320				 */
321				get_page(page); /* for putback_lru_page() */
322				__munlock_isolated_page(page);
323				unlock_page(page);
324				put_page(page); /* from follow_page_mask() */
325			}
326		}
327	}
328
 
 
 
 
 
 
 
329	/*
330	 * Phase 3: page putback for pages that qualified for the fast path
331	 * This will also call put_page() to return pin from follow_page_mask()
 
 
 
 
 
332	 */
333	if (pagevec_count(&pvec_putback))
334		__putback_lru_fast(&pvec_putback, pgrescued);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
335}
336
337/*
338 * Fill up pagevec for __munlock_pagevec using pte walk
339 *
340 * The function expects that the struct page corresponding to @start address is
341 * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone.
342 *
343 * The rest of @pvec is filled by subsequent pages within the same pmd and same
344 * zone, as long as the pte's are present and vm_normal_page() succeeds. These
345 * pages also get pinned.
346 *
347 * Returns the address of the next page that should be scanned. This equals
348 * @start + PAGE_SIZE when no page could be added by the pte walk.
349 */
350static unsigned long __munlock_pagevec_fill(struct pagevec *pvec,
351			struct vm_area_struct *vma, struct zone *zone,
352			unsigned long start, unsigned long end)
353{
354	pte_t *pte;
355	spinlock_t *ptl;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
356
357	/*
358	 * Initialize pte walk starting at the already pinned page where we
359	 * are sure that there is a pte, as it was pinned under the same
360	 * mmap_lock write op.
361	 */
362	pte = get_locked_pte(vma->vm_mm, start,	&ptl);
363	/* Make sure we do not cross the page table boundary */
364	end = pgd_addr_end(start, end);
365	end = p4d_addr_end(start, end);
366	end = pud_addr_end(start, end);
367	end = pmd_addr_end(start, end);
368
369	/* The page next to the pinned page is the first we will try to get */
370	start += PAGE_SIZE;
371	while (start < end) {
372		struct page *page = NULL;
373		pte++;
374		if (pte_present(*pte))
375			page = vm_normal_page(vma, start, *pte);
376		/*
377		 * Break if page could not be obtained or the page's node+zone does not
378		 * match
379		 */
380		if (!page || page_zone(page) != zone)
381			break;
382
383		/*
384		 * Do not use pagevec for PTE-mapped THP,
385		 * munlock_vma_pages_range() will handle them.
386		 */
387		if (PageTransCompound(page))
388			break;
 
389
390		get_page(page);
391		/*
392		 * Increase the address that will be returned *before* the
393		 * eventual break due to pvec becoming full by adding the page
394		 */
395		start += PAGE_SIZE;
396		if (pagevec_add(pvec, page) == 0)
397			break;
 
 
 
 
398	}
399	pte_unmap_unlock(pte, ptl);
400	return start;
 
 
 
401}
402
403/*
404 * munlock_vma_pages_range() - munlock all pages in the vma range.'
405 * @vma - vma containing range to be munlock()ed.
 
406 * @start - start address in @vma of the range
407 * @end - end of range in @vma.
408 *
409 *  For mremap(), munmap() and exit().
410 *
411 * Called with @vma VM_LOCKED.
412 *
413 * Returns with VM_LOCKED cleared.  Callers must be prepared to
414 * deal with this.
415 *
416 * We don't save and restore VM_LOCKED here because pages are
417 * still on lru.  In unmap path, pages might be scanned by reclaim
418 * and re-mlocked by page_mlock/try_to_unmap before we unmap and
419 * free them.  This will result in freeing mlocked pages.
420 */
421void munlock_vma_pages_range(struct vm_area_struct *vma,
422			     unsigned long start, unsigned long end)
423{
424	vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
425
426	while (start < end) {
427		struct page *page;
428		unsigned int page_mask = 0;
429		unsigned long page_increm;
430		struct pagevec pvec;
431		struct zone *zone;
432
433		pagevec_init(&pvec);
434		/*
435		 * Although FOLL_DUMP is intended for get_dump_page(),
436		 * it just so happens that its special treatment of the
437		 * ZERO_PAGE (returning an error instead of doing get_page)
438		 * suits munlock very well (and if somehow an abnormal page
439		 * has sneaked into the range, we won't oops here: great).
440		 */
441		page = follow_page(vma, start, FOLL_GET | FOLL_DUMP);
442
443		if (page && !IS_ERR(page)) {
444			if (PageTransTail(page)) {
445				VM_BUG_ON_PAGE(PageMlocked(page), page);
446				put_page(page); /* follow_page_mask() */
447			} else if (PageTransHuge(page)) {
448				lock_page(page);
449				/*
450				 * Any THP page found by follow_page_mask() may
451				 * have gotten split before reaching
452				 * munlock_vma_page(), so we need to compute
453				 * the page_mask here instead.
454				 */
455				page_mask = munlock_vma_page(page);
456				unlock_page(page);
457				put_page(page); /* follow_page_mask() */
458			} else {
459				/*
460				 * Non-huge pages are handled in batches via
461				 * pagevec. The pin from follow_page_mask()
462				 * prevents them from collapsing by THP.
463				 */
464				pagevec_add(&pvec, page);
465				zone = page_zone(page);
466
467				/*
468				 * Try to fill the rest of pagevec using fast
469				 * pte walk. This will also update start to
470				 * the next page to process. Then munlock the
471				 * pagevec.
472				 */
473				start = __munlock_pagevec_fill(&pvec, vma,
474						zone, start, end);
475				__munlock_pagevec(&pvec, zone);
476				goto next;
477			}
478		}
479		page_increm = 1 + page_mask;
480		start += page_increm * PAGE_SIZE;
481next:
482		cond_resched();
483	}
484}
485
486/*
487 * mlock_fixup  - handle mlock[all]/munlock[all] requests.
488 *
489 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
490 * munlock is a no-op.  However, for some special vmas, we go ahead and
491 * populate the ptes.
492 *
493 * For vmas that pass the filters, merge/split as appropriate.
494 */
495static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
496	unsigned long start, unsigned long end, vm_flags_t newflags)
 
497{
498	struct mm_struct *mm = vma->vm_mm;
499	pgoff_t pgoff;
500	int nr_pages;
501	int ret = 0;
502	int lock = !!(newflags & VM_LOCKED);
503	vm_flags_t old_flags = vma->vm_flags;
504
505	if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
506	    is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm) ||
507	    vma_is_dax(vma) || vma_is_secretmem(vma))
508		/* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
509		goto out;
510
511	pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
512	*prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
513			  vma->vm_file, pgoff, vma_policy(vma),
514			  vma->vm_userfaultfd_ctx);
515	if (*prev) {
516		vma = *prev;
517		goto success;
518	}
519
520	if (start != vma->vm_start) {
521		ret = split_vma(mm, vma, start, 1);
522		if (ret)
523			goto out;
524	}
525
526	if (end != vma->vm_end) {
527		ret = split_vma(mm, vma, end, 0);
528		if (ret)
529			goto out;
530	}
531
532success:
533	/*
534	 * Keep track of amount of locked VM.
535	 */
536	nr_pages = (end - start) >> PAGE_SHIFT;
537	if (!lock)
538		nr_pages = -nr_pages;
539	else if (old_flags & VM_LOCKED)
540		nr_pages = 0;
541	mm->locked_vm += nr_pages;
542
543	/*
544	 * vm_flags is protected by the mmap_lock held in write mode.
545	 * It's okay if try_to_unmap_one unmaps a page just after we
546	 * set VM_LOCKED, populate_vma_page_range will bring it back.
547	 */
548
549	if (lock)
550		vma->vm_flags = newflags;
551	else
552		munlock_vma_pages_range(vma, start, end);
553
 
554out:
555	*prev = vma;
556	return ret;
557}
558
559static int apply_vma_lock_flags(unsigned long start, size_t len,
560				vm_flags_t flags)
561{
562	unsigned long nstart, end, tmp;
563	struct vm_area_struct *vma, *prev;
564	int error;
565
566	VM_BUG_ON(offset_in_page(start));
567	VM_BUG_ON(len != PAGE_ALIGN(len));
568	end = start + len;
569	if (end < start)
570		return -EINVAL;
571	if (end == start)
572		return 0;
573	vma = find_vma(current->mm, start);
574	if (!vma || vma->vm_start > start)
575		return -ENOMEM;
576
577	prev = vma->vm_prev;
578	if (start > vma->vm_start)
579		prev = vma;
580
581	for (nstart = start ; ; ) {
582		vm_flags_t newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
 
 
 
583
584		newflags |= flags;
 
585
 
 
586		/* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
587		tmp = vma->vm_end;
588		if (tmp > end)
589			tmp = end;
590		error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
591		if (error)
592			break;
 
593		nstart = tmp;
594		if (nstart < prev->vm_end)
595			nstart = prev->vm_end;
596		if (nstart >= end)
597			break;
598
599		vma = prev->vm_next;
600		if (!vma || vma->vm_start != nstart) {
601			error = -ENOMEM;
602			break;
603		}
604	}
605	return error;
 
 
 
 
606}
607
608/*
609 * Go through vma areas and sum size of mlocked
610 * vma pages, as return value.
611 * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT)
612 * is also counted.
613 * Return value: previously mlocked page counts
614 */
615static unsigned long count_mm_mlocked_page_nr(struct mm_struct *mm,
616		unsigned long start, size_t len)
617{
618	struct vm_area_struct *vma;
619	unsigned long count = 0;
 
 
620
621	if (mm == NULL)
622		mm = current->mm;
623
624	vma = find_vma(mm, start);
625	if (vma == NULL)
626		return 0;
627
628	for (; vma ; vma = vma->vm_next) {
629		if (start >= vma->vm_end)
630			continue;
631		if (start + len <=  vma->vm_start)
632			break;
633		if (vma->vm_flags & VM_LOCKED) {
634			if (start > vma->vm_start)
635				count -= (start - vma->vm_start);
636			if (start + len < vma->vm_end) {
637				count += start + len - vma->vm_start;
638				break;
639			}
640			count += vma->vm_end - vma->vm_start;
641		}
642	}
643
644	return count >> PAGE_SHIFT;
645}
646
 
 
 
 
 
 
 
 
 
 
 
 
647static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
648{
649	unsigned long locked;
650	unsigned long lock_limit;
651	int error = -ENOMEM;
652
653	start = untagged_addr(start);
654
655	if (!can_do_mlock())
656		return -EPERM;
657
658	len = PAGE_ALIGN(len + (offset_in_page(start)));
659	start &= PAGE_MASK;
660
661	lock_limit = rlimit(RLIMIT_MEMLOCK);
662	lock_limit >>= PAGE_SHIFT;
663	locked = len >> PAGE_SHIFT;
664
665	if (mmap_write_lock_killable(current->mm))
666		return -EINTR;
667
668	locked += current->mm->locked_vm;
669	if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) {
670		/*
671		 * It is possible that the regions requested intersect with
672		 * previously mlocked areas, that part area in "mm->locked_vm"
673		 * should not be counted to new mlock increment count. So check
674		 * and adjust locked count if necessary.
675		 */
676		locked -= count_mm_mlocked_page_nr(current->mm,
677				start, len);
678	}
679
680	/* check against resource limits */
681	if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
682		error = apply_vma_lock_flags(start, len, flags);
683
684	mmap_write_unlock(current->mm);
685	if (error)
686		return error;
687
688	error = __mm_populate(start, len, 0);
689	if (error)
690		return __mlock_posix_error_return(error);
691	return 0;
692}
693
694SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
695{
696	return do_mlock(start, len, VM_LOCKED);
697}
698
699SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
700{
701	vm_flags_t vm_flags = VM_LOCKED;
702
703	if (flags & ~MLOCK_ONFAULT)
704		return -EINVAL;
705
706	if (flags & MLOCK_ONFAULT)
707		vm_flags |= VM_LOCKONFAULT;
708
709	return do_mlock(start, len, vm_flags);
710}
711
712SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
713{
714	int ret;
715
716	start = untagged_addr(start);
717
718	len = PAGE_ALIGN(len + (offset_in_page(start)));
719	start &= PAGE_MASK;
720
721	if (mmap_write_lock_killable(current->mm))
722		return -EINTR;
723	ret = apply_vma_lock_flags(start, len, 0);
724	mmap_write_unlock(current->mm);
725
726	return ret;
727}
728
729/*
730 * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
731 * and translate into the appropriate modifications to mm->def_flags and/or the
732 * flags for all current VMAs.
733 *
734 * There are a couple of subtleties with this.  If mlockall() is called multiple
735 * times with different flags, the values do not necessarily stack.  If mlockall
736 * is called once including the MCL_FUTURE flag and then a second time without
737 * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
738 */
739static int apply_mlockall_flags(int flags)
740{
 
741	struct vm_area_struct *vma, *prev = NULL;
742	vm_flags_t to_add = 0;
743
744	current->mm->def_flags &= VM_LOCKED_CLEAR_MASK;
745	if (flags & MCL_FUTURE) {
746		current->mm->def_flags |= VM_LOCKED;
747
748		if (flags & MCL_ONFAULT)
749			current->mm->def_flags |= VM_LOCKONFAULT;
750
751		if (!(flags & MCL_CURRENT))
752			goto out;
753	}
754
755	if (flags & MCL_CURRENT) {
756		to_add |= VM_LOCKED;
757		if (flags & MCL_ONFAULT)
758			to_add |= VM_LOCKONFAULT;
759	}
760
761	for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
762		vm_flags_t newflags;
763
764		newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
765		newflags |= to_add;
766
767		/* Ignore errors */
768		mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
 
769		cond_resched();
770	}
771out:
772	return 0;
773}
774
775SYSCALL_DEFINE1(mlockall, int, flags)
776{
777	unsigned long lock_limit;
778	int ret;
779
780	if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)) ||
781	    flags == MCL_ONFAULT)
782		return -EINVAL;
783
784	if (!can_do_mlock())
785		return -EPERM;
786
787	lock_limit = rlimit(RLIMIT_MEMLOCK);
788	lock_limit >>= PAGE_SHIFT;
789
790	if (mmap_write_lock_killable(current->mm))
791		return -EINTR;
792
793	ret = -ENOMEM;
794	if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
795	    capable(CAP_IPC_LOCK))
796		ret = apply_mlockall_flags(flags);
797	mmap_write_unlock(current->mm);
798	if (!ret && (flags & MCL_CURRENT))
799		mm_populate(0, TASK_SIZE);
800
801	return ret;
802}
803
804SYSCALL_DEFINE0(munlockall)
805{
806	int ret;
807
808	if (mmap_write_lock_killable(current->mm))
809		return -EINTR;
810	ret = apply_mlockall_flags(0);
811	mmap_write_unlock(current->mm);
812	return ret;
813}
814
815/*
816 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
817 * shm segments) get accounted against the user_struct instead.
818 */
819static DEFINE_SPINLOCK(shmlock_user_lock);
820
821int user_shm_lock(size_t size, struct ucounts *ucounts)
822{
823	unsigned long lock_limit, locked;
824	long memlock;
825	int allowed = 0;
826
827	locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
828	lock_limit = rlimit(RLIMIT_MEMLOCK);
829	if (lock_limit == RLIM_INFINITY)
830		allowed = 1;
831	lock_limit >>= PAGE_SHIFT;
832	spin_lock(&shmlock_user_lock);
833	memlock = inc_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
834
835	if (!allowed && (memlock == LONG_MAX || memlock > lock_limit) && !capable(CAP_IPC_LOCK)) {
836		dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
837		goto out;
838	}
839	if (!get_ucounts(ucounts)) {
840		dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
 
841		goto out;
842	}
843	allowed = 1;
844out:
845	spin_unlock(&shmlock_user_lock);
846	return allowed;
847}
848
849void user_shm_unlock(size_t size, struct ucounts *ucounts)
850{
851	spin_lock(&shmlock_user_lock);
852	dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, (size + PAGE_SIZE - 1) >> PAGE_SHIFT);
853	spin_unlock(&shmlock_user_lock);
854	put_ucounts(ucounts);
855}
v6.9.4
  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_fbatch {
 32	local_lock_t lock;
 33	struct folio_batch fbatch;
 34};
 35
 36static DEFINE_PER_CPU(struct mlock_fbatch, mlock_fbatch) = {
 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 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
 61static 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));
 98out:
 99	folio_set_lru(folio);
100	return lruvec;
101}
102
103static 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));
116out:
117	lruvec_add_folio(lruvec, folio);
118	folio_set_lru(folio);
119	return lruvec;
120}
121
122static 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
142munlock:
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	}
158out:
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
169static inline struct folio *mlock_lru(struct folio *folio)
170{
171	return (struct folio *)((unsigned long)folio + LRU_FOLIO);
172}
173
174static 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 */
186static 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
212void 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
223void 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
233bool 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 */
242void 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 */
267void 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 */
290void 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
307static 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
332static 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
363static 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 = page_folio(pmd_page(*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
412next_entry:
413		pte += step - 1;
414		addr += (step - 1) << PAGE_SHIFT;
415	}
416	pte_unmap(start_pte);
417out:
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 */
434static 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 */
477static 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	}
520out:
521	*prev = vma;
522	return ret;
523}
524
525static 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 */
582static 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 */
614static 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
623static __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
670SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
671{
672	return do_mlock(start, len, VM_LOCKED);
673}
674
675SYSCALL_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
688SYSCALL_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 */
715static 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	}
749out:
750	return 0;
751}
752
753SYSCALL_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
782SYSCALL_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 */
797static DEFINE_SPINLOCK(shmlock_user_lock);
798
799int 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;
822out:
823	spin_unlock(&shmlock_user_lock);
824	return allowed;
825}
826
827void 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}