xref: /openbmc/linux/mm/mlock.c (revision 930beb5a)
1 /*
2  *	linux/mm/mlock.c
3  *
4  *  (C) Copyright 1995 Linus Torvalds
5  *  (C) Copyright 2002 Christoph Hellwig
6  */
7 
8 #include <linux/capability.h>
9 #include <linux/mman.h>
10 #include <linux/mm.h>
11 #include <linux/swap.h>
12 #include <linux/swapops.h>
13 #include <linux/pagemap.h>
14 #include <linux/pagevec.h>
15 #include <linux/mempolicy.h>
16 #include <linux/syscalls.h>
17 #include <linux/sched.h>
18 #include <linux/export.h>
19 #include <linux/rmap.h>
20 #include <linux/mmzone.h>
21 #include <linux/hugetlb.h>
22 #include <linux/memcontrol.h>
23 #include <linux/mm_inline.h>
24 
25 #include "internal.h"
26 
27 int can_do_mlock(void)
28 {
29 	if (capable(CAP_IPC_LOCK))
30 		return 1;
31 	if (rlimit(RLIMIT_MEMLOCK) != 0)
32 		return 1;
33 	return 0;
34 }
35 EXPORT_SYMBOL(can_do_mlock);
36 
37 /*
38  * Mlocked pages are marked with PageMlocked() flag for efficient testing
39  * in vmscan and, possibly, the fault path; and to support semi-accurate
40  * statistics.
41  *
42  * An mlocked page [PageMlocked(page)] is unevictable.  As such, it will
43  * be placed on the LRU "unevictable" list, rather than the [in]active lists.
44  * The unevictable list is an LRU sibling list to the [in]active lists.
45  * PageUnevictable is set to indicate the unevictable state.
46  *
47  * When lazy mlocking via vmscan, it is important to ensure that the
48  * vma's VM_LOCKED status is not concurrently being modified, otherwise we
49  * may have mlocked a page that is being munlocked. So lazy mlock must take
50  * the mmap_sem for read, and verify that the vma really is locked
51  * (see mm/rmap.c).
52  */
53 
54 /*
55  *  LRU accounting for clear_page_mlock()
56  */
57 void clear_page_mlock(struct page *page)
58 {
59 	if (!TestClearPageMlocked(page))
60 		return;
61 
62 	mod_zone_page_state(page_zone(page), NR_MLOCK,
63 			    -hpage_nr_pages(page));
64 	count_vm_event(UNEVICTABLE_PGCLEARED);
65 	if (!isolate_lru_page(page)) {
66 		putback_lru_page(page);
67 	} else {
68 		/*
69 		 * We lost the race. the page already moved to evictable list.
70 		 */
71 		if (PageUnevictable(page))
72 			count_vm_event(UNEVICTABLE_PGSTRANDED);
73 	}
74 }
75 
76 /*
77  * Mark page as mlocked if not already.
78  * If page on LRU, isolate and putback to move to unevictable list.
79  */
80 void mlock_vma_page(struct page *page)
81 {
82 	BUG_ON(!PageLocked(page));
83 
84 	if (!TestSetPageMlocked(page)) {
85 		mod_zone_page_state(page_zone(page), NR_MLOCK,
86 				    hpage_nr_pages(page));
87 		count_vm_event(UNEVICTABLE_PGMLOCKED);
88 		if (!isolate_lru_page(page))
89 			putback_lru_page(page);
90 	}
91 }
92 
93 /*
94  * Finish munlock after successful page isolation
95  *
96  * Page must be locked. This is a wrapper for try_to_munlock()
97  * and putback_lru_page() with munlock accounting.
98  */
99 static void __munlock_isolated_page(struct page *page)
100 {
101 	int ret = SWAP_AGAIN;
102 
103 	/*
104 	 * Optimization: if the page was mapped just once, that's our mapping
105 	 * and we don't need to check all the other vmas.
106 	 */
107 	if (page_mapcount(page) > 1)
108 		ret = try_to_munlock(page);
109 
110 	/* Did try_to_unlock() succeed or punt? */
111 	if (ret != SWAP_MLOCK)
112 		count_vm_event(UNEVICTABLE_PGMUNLOCKED);
113 
114 	putback_lru_page(page);
115 }
116 
117 /*
118  * Accounting for page isolation fail during munlock
119  *
120  * Performs accounting when page isolation fails in munlock. There is nothing
121  * else to do because it means some other task has already removed the page
122  * from the LRU. putback_lru_page() will take care of removing the page from
123  * the unevictable list, if necessary. vmscan [page_referenced()] will move
124  * the page back to the unevictable list if some other vma has it mlocked.
125  */
126 static void __munlock_isolation_failed(struct page *page)
127 {
128 	if (PageUnevictable(page))
129 		count_vm_event(UNEVICTABLE_PGSTRANDED);
130 	else
131 		count_vm_event(UNEVICTABLE_PGMUNLOCKED);
132 }
133 
134 /**
135  * munlock_vma_page - munlock a vma page
136  * @page - page to be unlocked, either a normal page or THP page head
137  *
138  * returns the size of the page as a page mask (0 for normal page,
139  *         HPAGE_PMD_NR - 1 for THP head page)
140  *
141  * called from munlock()/munmap() path with page supposedly on the LRU.
142  * When we munlock a page, because the vma where we found the page is being
143  * munlock()ed or munmap()ed, we want to check whether other vmas hold the
144  * page locked so that we can leave it on the unevictable lru list and not
145  * bother vmscan with it.  However, to walk the page's rmap list in
146  * try_to_munlock() we must isolate the page from the LRU.  If some other
147  * task has removed the page from the LRU, we won't be able to do that.
148  * So we clear the PageMlocked as we might not get another chance.  If we
149  * can't isolate the page, we leave it for putback_lru_page() and vmscan
150  * [page_referenced()/try_to_unmap()] to deal with.
151  */
152 unsigned int munlock_vma_page(struct page *page)
153 {
154 	unsigned int nr_pages;
155 
156 	BUG_ON(!PageLocked(page));
157 
158 	if (TestClearPageMlocked(page)) {
159 		nr_pages = hpage_nr_pages(page);
160 		mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
161 		if (!isolate_lru_page(page))
162 			__munlock_isolated_page(page);
163 		else
164 			__munlock_isolation_failed(page);
165 	} else {
166 		nr_pages = hpage_nr_pages(page);
167 	}
168 
169 	/*
170 	 * Regardless of the original PageMlocked flag, we determine nr_pages
171 	 * after touching the flag. This leaves a possible race with a THP page
172 	 * split, such that a whole THP page was munlocked, but nr_pages == 1.
173 	 * Returning a smaller mask due to that is OK, the worst that can
174 	 * happen is subsequent useless scanning of the former tail pages.
175 	 * The NR_MLOCK accounting can however become broken.
176 	 */
177 	return nr_pages - 1;
178 }
179 
180 /**
181  * __mlock_vma_pages_range() -  mlock a range of pages in the vma.
182  * @vma:   target vma
183  * @start: start address
184  * @end:   end address
185  *
186  * This takes care of making the pages present too.
187  *
188  * return 0 on success, negative error code on error.
189  *
190  * vma->vm_mm->mmap_sem must be held for at least read.
191  */
192 long __mlock_vma_pages_range(struct vm_area_struct *vma,
193 		unsigned long start, unsigned long end, int *nonblocking)
194 {
195 	struct mm_struct *mm = vma->vm_mm;
196 	unsigned long nr_pages = (end - start) / PAGE_SIZE;
197 	int gup_flags;
198 
199 	VM_BUG_ON(start & ~PAGE_MASK);
200 	VM_BUG_ON(end   & ~PAGE_MASK);
201 	VM_BUG_ON(start < vma->vm_start);
202 	VM_BUG_ON(end   > vma->vm_end);
203 	VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem));
204 
205 	gup_flags = FOLL_TOUCH | FOLL_MLOCK;
206 	/*
207 	 * We want to touch writable mappings with a write fault in order
208 	 * to break COW, except for shared mappings because these don't COW
209 	 * and we would not want to dirty them for nothing.
210 	 */
211 	if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
212 		gup_flags |= FOLL_WRITE;
213 
214 	/*
215 	 * We want mlock to succeed for regions that have any permissions
216 	 * other than PROT_NONE.
217 	 */
218 	if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))
219 		gup_flags |= FOLL_FORCE;
220 
221 	/*
222 	 * We made sure addr is within a VMA, so the following will
223 	 * not result in a stack expansion that recurses back here.
224 	 */
225 	return __get_user_pages(current, mm, start, nr_pages, gup_flags,
226 				NULL, NULL, nonblocking);
227 }
228 
229 /*
230  * convert get_user_pages() return value to posix mlock() error
231  */
232 static int __mlock_posix_error_return(long retval)
233 {
234 	if (retval == -EFAULT)
235 		retval = -ENOMEM;
236 	else if (retval == -ENOMEM)
237 		retval = -EAGAIN;
238 	return retval;
239 }
240 
241 /*
242  * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec()
243  *
244  * The fast path is available only for evictable pages with single mapping.
245  * Then we can bypass the per-cpu pvec and get better performance.
246  * when mapcount > 1 we need try_to_munlock() which can fail.
247  * when !page_evictable(), we need the full redo logic of putback_lru_page to
248  * avoid leaving evictable page in unevictable list.
249  *
250  * In case of success, @page is added to @pvec and @pgrescued is incremented
251  * in case that the page was previously unevictable. @page is also unlocked.
252  */
253 static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec,
254 		int *pgrescued)
255 {
256 	VM_BUG_ON(PageLRU(page));
257 	VM_BUG_ON(!PageLocked(page));
258 
259 	if (page_mapcount(page) <= 1 && page_evictable(page)) {
260 		pagevec_add(pvec, page);
261 		if (TestClearPageUnevictable(page))
262 			(*pgrescued)++;
263 		unlock_page(page);
264 		return true;
265 	}
266 
267 	return false;
268 }
269 
270 /*
271  * Putback multiple evictable pages to the LRU
272  *
273  * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of
274  * the pages might have meanwhile become unevictable but that is OK.
275  */
276 static void __putback_lru_fast(struct pagevec *pvec, int pgrescued)
277 {
278 	count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec));
279 	/*
280 	 *__pagevec_lru_add() calls release_pages() so we don't call
281 	 * put_page() explicitly
282 	 */
283 	__pagevec_lru_add(pvec);
284 	count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued);
285 }
286 
287 /*
288  * Munlock a batch of pages from the same zone
289  *
290  * The work is split to two main phases. First phase clears the Mlocked flag
291  * and attempts to isolate the pages, all under a single zone lru lock.
292  * The second phase finishes the munlock only for pages where isolation
293  * succeeded.
294  *
295  * Note that the pagevec may be modified during the process.
296  */
297 static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone)
298 {
299 	int i;
300 	int nr = pagevec_count(pvec);
301 	int delta_munlocked;
302 	struct pagevec pvec_putback;
303 	int pgrescued = 0;
304 
305 	pagevec_init(&pvec_putback, 0);
306 
307 	/* Phase 1: page isolation */
308 	spin_lock_irq(&zone->lru_lock);
309 	for (i = 0; i < nr; i++) {
310 		struct page *page = pvec->pages[i];
311 
312 		if (TestClearPageMlocked(page)) {
313 			struct lruvec *lruvec;
314 			int lru;
315 
316 			if (PageLRU(page)) {
317 				lruvec = mem_cgroup_page_lruvec(page, zone);
318 				lru = page_lru(page);
319 				/*
320 				 * We already have pin from follow_page_mask()
321 				 * so we can spare the get_page() here.
322 				 */
323 				ClearPageLRU(page);
324 				del_page_from_lru_list(page, lruvec, lru);
325 			} else {
326 				__munlock_isolation_failed(page);
327 				goto skip_munlock;
328 			}
329 
330 		} else {
331 skip_munlock:
332 			/*
333 			 * We won't be munlocking this page in the next phase
334 			 * but we still need to release the follow_page_mask()
335 			 * pin. We cannot do it under lru_lock however. If it's
336 			 * the last pin, __page_cache_release would deadlock.
337 			 */
338 			pagevec_add(&pvec_putback, pvec->pages[i]);
339 			pvec->pages[i] = NULL;
340 		}
341 	}
342 	delta_munlocked = -nr + pagevec_count(&pvec_putback);
343 	__mod_zone_page_state(zone, NR_MLOCK, delta_munlocked);
344 	spin_unlock_irq(&zone->lru_lock);
345 
346 	/* Now we can release pins of pages that we are not munlocking */
347 	pagevec_release(&pvec_putback);
348 
349 	/* Phase 2: page munlock */
350 	for (i = 0; i < nr; i++) {
351 		struct page *page = pvec->pages[i];
352 
353 		if (page) {
354 			lock_page(page);
355 			if (!__putback_lru_fast_prepare(page, &pvec_putback,
356 					&pgrescued)) {
357 				/*
358 				 * Slow path. We don't want to lose the last
359 				 * pin before unlock_page()
360 				 */
361 				get_page(page); /* for putback_lru_page() */
362 				__munlock_isolated_page(page);
363 				unlock_page(page);
364 				put_page(page); /* from follow_page_mask() */
365 			}
366 		}
367 	}
368 
369 	/*
370 	 * Phase 3: page putback for pages that qualified for the fast path
371 	 * This will also call put_page() to return pin from follow_page_mask()
372 	 */
373 	if (pagevec_count(&pvec_putback))
374 		__putback_lru_fast(&pvec_putback, pgrescued);
375 }
376 
377 /*
378  * Fill up pagevec for __munlock_pagevec using pte walk
379  *
380  * The function expects that the struct page corresponding to @start address is
381  * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone.
382  *
383  * The rest of @pvec is filled by subsequent pages within the same pmd and same
384  * zone, as long as the pte's are present and vm_normal_page() succeeds. These
385  * pages also get pinned.
386  *
387  * Returns the address of the next page that should be scanned. This equals
388  * @start + PAGE_SIZE when no page could be added by the pte walk.
389  */
390 static unsigned long __munlock_pagevec_fill(struct pagevec *pvec,
391 		struct vm_area_struct *vma, int zoneid,	unsigned long start,
392 		unsigned long end)
393 {
394 	pte_t *pte;
395 	spinlock_t *ptl;
396 
397 	/*
398 	 * Initialize pte walk starting at the already pinned page where we
399 	 * are sure that there is a pte, as it was pinned under the same
400 	 * mmap_sem write op.
401 	 */
402 	pte = get_locked_pte(vma->vm_mm, start,	&ptl);
403 	/* Make sure we do not cross the page table boundary */
404 	end = pgd_addr_end(start, end);
405 	end = pud_addr_end(start, end);
406 	end = pmd_addr_end(start, end);
407 
408 	/* The page next to the pinned page is the first we will try to get */
409 	start += PAGE_SIZE;
410 	while (start < end) {
411 		struct page *page = NULL;
412 		pte++;
413 		if (pte_present(*pte))
414 			page = vm_normal_page(vma, start, *pte);
415 		/*
416 		 * Break if page could not be obtained or the page's node+zone does not
417 		 * match
418 		 */
419 		if (!page || page_zone_id(page) != zoneid)
420 			break;
421 
422 		get_page(page);
423 		/*
424 		 * Increase the address that will be returned *before* the
425 		 * eventual break due to pvec becoming full by adding the page
426 		 */
427 		start += PAGE_SIZE;
428 		if (pagevec_add(pvec, page) == 0)
429 			break;
430 	}
431 	pte_unmap_unlock(pte, ptl);
432 	return start;
433 }
434 
435 /*
436  * munlock_vma_pages_range() - munlock all pages in the vma range.'
437  * @vma - vma containing range to be munlock()ed.
438  * @start - start address in @vma of the range
439  * @end - end of range in @vma.
440  *
441  *  For mremap(), munmap() and exit().
442  *
443  * Called with @vma VM_LOCKED.
444  *
445  * Returns with VM_LOCKED cleared.  Callers must be prepared to
446  * deal with this.
447  *
448  * We don't save and restore VM_LOCKED here because pages are
449  * still on lru.  In unmap path, pages might be scanned by reclaim
450  * and re-mlocked by try_to_{munlock|unmap} before we unmap and
451  * free them.  This will result in freeing mlocked pages.
452  */
453 void munlock_vma_pages_range(struct vm_area_struct *vma,
454 			     unsigned long start, unsigned long end)
455 {
456 	vma->vm_flags &= ~VM_LOCKED;
457 
458 	while (start < end) {
459 		struct page *page = NULL;
460 		unsigned int page_mask;
461 		unsigned long page_increm;
462 		struct pagevec pvec;
463 		struct zone *zone;
464 		int zoneid;
465 
466 		pagevec_init(&pvec, 0);
467 		/*
468 		 * Although FOLL_DUMP is intended for get_dump_page(),
469 		 * it just so happens that its special treatment of the
470 		 * ZERO_PAGE (returning an error instead of doing get_page)
471 		 * suits munlock very well (and if somehow an abnormal page
472 		 * has sneaked into the range, we won't oops here: great).
473 		 */
474 		page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP,
475 				&page_mask);
476 
477 		if (page && !IS_ERR(page)) {
478 			if (PageTransHuge(page)) {
479 				lock_page(page);
480 				/*
481 				 * Any THP page found by follow_page_mask() may
482 				 * have gotten split before reaching
483 				 * munlock_vma_page(), so we need to recompute
484 				 * the page_mask here.
485 				 */
486 				page_mask = munlock_vma_page(page);
487 				unlock_page(page);
488 				put_page(page); /* follow_page_mask() */
489 			} else {
490 				/*
491 				 * Non-huge pages are handled in batches via
492 				 * pagevec. The pin from follow_page_mask()
493 				 * prevents them from collapsing by THP.
494 				 */
495 				pagevec_add(&pvec, page);
496 				zone = page_zone(page);
497 				zoneid = page_zone_id(page);
498 
499 				/*
500 				 * Try to fill the rest of pagevec using fast
501 				 * pte walk. This will also update start to
502 				 * the next page to process. Then munlock the
503 				 * pagevec.
504 				 */
505 				start = __munlock_pagevec_fill(&pvec, vma,
506 						zoneid, start, end);
507 				__munlock_pagevec(&pvec, zone);
508 				goto next;
509 			}
510 		}
511 		/* It's a bug to munlock in the middle of a THP page */
512 		VM_BUG_ON((start >> PAGE_SHIFT) & page_mask);
513 		page_increm = 1 + page_mask;
514 		start += page_increm * PAGE_SIZE;
515 next:
516 		cond_resched();
517 	}
518 }
519 
520 /*
521  * mlock_fixup  - handle mlock[all]/munlock[all] requests.
522  *
523  * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
524  * munlock is a no-op.  However, for some special vmas, we go ahead and
525  * populate the ptes.
526  *
527  * For vmas that pass the filters, merge/split as appropriate.
528  */
529 static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
530 	unsigned long start, unsigned long end, vm_flags_t newflags)
531 {
532 	struct mm_struct *mm = vma->vm_mm;
533 	pgoff_t pgoff;
534 	int nr_pages;
535 	int ret = 0;
536 	int lock = !!(newflags & VM_LOCKED);
537 
538 	if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
539 	    is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
540 		goto out;	/* don't set VM_LOCKED,  don't count */
541 
542 	pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
543 	*prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
544 			  vma->vm_file, pgoff, vma_policy(vma));
545 	if (*prev) {
546 		vma = *prev;
547 		goto success;
548 	}
549 
550 	if (start != vma->vm_start) {
551 		ret = split_vma(mm, vma, start, 1);
552 		if (ret)
553 			goto out;
554 	}
555 
556 	if (end != vma->vm_end) {
557 		ret = split_vma(mm, vma, end, 0);
558 		if (ret)
559 			goto out;
560 	}
561 
562 success:
563 	/*
564 	 * Keep track of amount of locked VM.
565 	 */
566 	nr_pages = (end - start) >> PAGE_SHIFT;
567 	if (!lock)
568 		nr_pages = -nr_pages;
569 	mm->locked_vm += nr_pages;
570 
571 	/*
572 	 * vm_flags is protected by the mmap_sem held in write mode.
573 	 * It's okay if try_to_unmap_one unmaps a page just after we
574 	 * set VM_LOCKED, __mlock_vma_pages_range will bring it back.
575 	 */
576 
577 	if (lock)
578 		vma->vm_flags = newflags;
579 	else
580 		munlock_vma_pages_range(vma, start, end);
581 
582 out:
583 	*prev = vma;
584 	return ret;
585 }
586 
587 static int do_mlock(unsigned long start, size_t len, int on)
588 {
589 	unsigned long nstart, end, tmp;
590 	struct vm_area_struct * vma, * prev;
591 	int error;
592 
593 	VM_BUG_ON(start & ~PAGE_MASK);
594 	VM_BUG_ON(len != PAGE_ALIGN(len));
595 	end = start + len;
596 	if (end < start)
597 		return -EINVAL;
598 	if (end == start)
599 		return 0;
600 	vma = find_vma(current->mm, start);
601 	if (!vma || vma->vm_start > start)
602 		return -ENOMEM;
603 
604 	prev = vma->vm_prev;
605 	if (start > vma->vm_start)
606 		prev = vma;
607 
608 	for (nstart = start ; ; ) {
609 		vm_flags_t newflags;
610 
611 		/* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
612 
613 		newflags = vma->vm_flags & ~VM_LOCKED;
614 		if (on)
615 			newflags |= VM_LOCKED;
616 
617 		tmp = vma->vm_end;
618 		if (tmp > end)
619 			tmp = end;
620 		error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
621 		if (error)
622 			break;
623 		nstart = tmp;
624 		if (nstart < prev->vm_end)
625 			nstart = prev->vm_end;
626 		if (nstart >= end)
627 			break;
628 
629 		vma = prev->vm_next;
630 		if (!vma || vma->vm_start != nstart) {
631 			error = -ENOMEM;
632 			break;
633 		}
634 	}
635 	return error;
636 }
637 
638 /*
639  * __mm_populate - populate and/or mlock pages within a range of address space.
640  *
641  * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap
642  * flags. VMAs must be already marked with the desired vm_flags, and
643  * mmap_sem must not be held.
644  */
645 int __mm_populate(unsigned long start, unsigned long len, int ignore_errors)
646 {
647 	struct mm_struct *mm = current->mm;
648 	unsigned long end, nstart, nend;
649 	struct vm_area_struct *vma = NULL;
650 	int locked = 0;
651 	long ret = 0;
652 
653 	VM_BUG_ON(start & ~PAGE_MASK);
654 	VM_BUG_ON(len != PAGE_ALIGN(len));
655 	end = start + len;
656 
657 	for (nstart = start; nstart < end; nstart = nend) {
658 		/*
659 		 * We want to fault in pages for [nstart; end) address range.
660 		 * Find first corresponding VMA.
661 		 */
662 		if (!locked) {
663 			locked = 1;
664 			down_read(&mm->mmap_sem);
665 			vma = find_vma(mm, nstart);
666 		} else if (nstart >= vma->vm_end)
667 			vma = vma->vm_next;
668 		if (!vma || vma->vm_start >= end)
669 			break;
670 		/*
671 		 * Set [nstart; nend) to intersection of desired address
672 		 * range with the first VMA. Also, skip undesirable VMA types.
673 		 */
674 		nend = min(end, vma->vm_end);
675 		if (vma->vm_flags & (VM_IO | VM_PFNMAP))
676 			continue;
677 		if (nstart < vma->vm_start)
678 			nstart = vma->vm_start;
679 		/*
680 		 * Now fault in a range of pages. __mlock_vma_pages_range()
681 		 * double checks the vma flags, so that it won't mlock pages
682 		 * if the vma was already munlocked.
683 		 */
684 		ret = __mlock_vma_pages_range(vma, nstart, nend, &locked);
685 		if (ret < 0) {
686 			if (ignore_errors) {
687 				ret = 0;
688 				continue;	/* continue at next VMA */
689 			}
690 			ret = __mlock_posix_error_return(ret);
691 			break;
692 		}
693 		nend = nstart + ret * PAGE_SIZE;
694 		ret = 0;
695 	}
696 	if (locked)
697 		up_read(&mm->mmap_sem);
698 	return ret;	/* 0 or negative error code */
699 }
700 
701 SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
702 {
703 	unsigned long locked;
704 	unsigned long lock_limit;
705 	int error = -ENOMEM;
706 
707 	if (!can_do_mlock())
708 		return -EPERM;
709 
710 	lru_add_drain_all();	/* flush pagevec */
711 
712 	down_write(&current->mm->mmap_sem);
713 	len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
714 	start &= PAGE_MASK;
715 
716 	locked = len >> PAGE_SHIFT;
717 	locked += current->mm->locked_vm;
718 
719 	lock_limit = rlimit(RLIMIT_MEMLOCK);
720 	lock_limit >>= PAGE_SHIFT;
721 
722 	/* check against resource limits */
723 	if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
724 		error = do_mlock(start, len, 1);
725 	up_write(&current->mm->mmap_sem);
726 	if (!error)
727 		error = __mm_populate(start, len, 0);
728 	return error;
729 }
730 
731 SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
732 {
733 	int ret;
734 
735 	down_write(&current->mm->mmap_sem);
736 	len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
737 	start &= PAGE_MASK;
738 	ret = do_mlock(start, len, 0);
739 	up_write(&current->mm->mmap_sem);
740 	return ret;
741 }
742 
743 static int do_mlockall(int flags)
744 {
745 	struct vm_area_struct * vma, * prev = NULL;
746 
747 	if (flags & MCL_FUTURE)
748 		current->mm->def_flags |= VM_LOCKED;
749 	else
750 		current->mm->def_flags &= ~VM_LOCKED;
751 	if (flags == MCL_FUTURE)
752 		goto out;
753 
754 	for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
755 		vm_flags_t newflags;
756 
757 		newflags = vma->vm_flags & ~VM_LOCKED;
758 		if (flags & MCL_CURRENT)
759 			newflags |= VM_LOCKED;
760 
761 		/* Ignore errors */
762 		mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
763 		cond_resched();
764 	}
765 out:
766 	return 0;
767 }
768 
769 SYSCALL_DEFINE1(mlockall, int, flags)
770 {
771 	unsigned long lock_limit;
772 	int ret = -EINVAL;
773 
774 	if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
775 		goto out;
776 
777 	ret = -EPERM;
778 	if (!can_do_mlock())
779 		goto out;
780 
781 	if (flags & MCL_CURRENT)
782 		lru_add_drain_all();	/* flush pagevec */
783 
784 	down_write(&current->mm->mmap_sem);
785 
786 	lock_limit = rlimit(RLIMIT_MEMLOCK);
787 	lock_limit >>= PAGE_SHIFT;
788 
789 	ret = -ENOMEM;
790 	if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
791 	    capable(CAP_IPC_LOCK))
792 		ret = do_mlockall(flags);
793 	up_write(&current->mm->mmap_sem);
794 	if (!ret && (flags & MCL_CURRENT))
795 		mm_populate(0, TASK_SIZE);
796 out:
797 	return ret;
798 }
799 
800 SYSCALL_DEFINE0(munlockall)
801 {
802 	int ret;
803 
804 	down_write(&current->mm->mmap_sem);
805 	ret = do_mlockall(0);
806 	up_write(&current->mm->mmap_sem);
807 	return ret;
808 }
809 
810 /*
811  * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
812  * shm segments) get accounted against the user_struct instead.
813  */
814 static DEFINE_SPINLOCK(shmlock_user_lock);
815 
816 int user_shm_lock(size_t size, struct user_struct *user)
817 {
818 	unsigned long lock_limit, locked;
819 	int allowed = 0;
820 
821 	locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
822 	lock_limit = rlimit(RLIMIT_MEMLOCK);
823 	if (lock_limit == RLIM_INFINITY)
824 		allowed = 1;
825 	lock_limit >>= PAGE_SHIFT;
826 	spin_lock(&shmlock_user_lock);
827 	if (!allowed &&
828 	    locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
829 		goto out;
830 	get_uid(user);
831 	user->locked_shm += locked;
832 	allowed = 1;
833 out:
834 	spin_unlock(&shmlock_user_lock);
835 	return allowed;
836 }
837 
838 void user_shm_unlock(size_t size, struct user_struct *user)
839 {
840 	spin_lock(&shmlock_user_lock);
841 	user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
842 	spin_unlock(&shmlock_user_lock);
843 	free_uid(user);
844 }
845