4 * (C) Copyright 1995 Linus Torvalds
5 * (C) Copyright 2002 Christoph Hellwig
8 #include <linux/capability.h>
9 #include <linux/mman.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>
27 int can_do_mlock(void)
29 if (capable(CAP_IPC_LOCK))
31 if (rlimit(RLIMIT_MEMLOCK) != 0)
35 EXPORT_SYMBOL(can_do_mlock);
38 * Mlocked pages are marked with PageMlocked() flag for efficient testing
39 * in vmscan and, possibly, the fault path; and to support semi-accurate
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.
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
55 * LRU accounting for clear_page_mlock()
57 void clear_page_mlock(struct page *page)
59 if (!TestClearPageMlocked(page))
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);
69 * We lost the race. the page already moved to evictable list.
71 if (PageUnevictable(page))
72 count_vm_event(UNEVICTABLE_PGSTRANDED);
77 * Mark page as mlocked if not already.
78 * If page on LRU, isolate and putback to move to unevictable list.
80 void mlock_vma_page(struct page *page)
82 BUG_ON(!PageLocked(page));
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);
94 * Isolate a page from LRU with optional get_page() pin.
95 * Assumes lru_lock already held and page already pinned.
97 static bool __munlock_isolate_lru_page(struct page *page, bool getpage)
100 struct lruvec *lruvec;
102 lruvec = mem_cgroup_page_lruvec(page, page_zone(page));
106 del_page_from_lru_list(page, lruvec, page_lru(page));
114 * Finish munlock after successful page isolation
116 * Page must be locked. This is a wrapper for try_to_munlock()
117 * and putback_lru_page() with munlock accounting.
119 static void __munlock_isolated_page(struct page *page)
121 int ret = SWAP_AGAIN;
124 * Optimization: if the page was mapped just once, that's our mapping
125 * and we don't need to check all the other vmas.
127 if (page_mapcount(page) > 1)
128 ret = try_to_munlock(page);
130 /* Did try_to_unlock() succeed or punt? */
131 if (ret != SWAP_MLOCK)
132 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
134 putback_lru_page(page);
138 * Accounting for page isolation fail during munlock
140 * Performs accounting when page isolation fails in munlock. There is nothing
141 * else to do because it means some other task has already removed the page
142 * from the LRU. putback_lru_page() will take care of removing the page from
143 * the unevictable list, if necessary. vmscan [page_referenced()] will move
144 * the page back to the unevictable list if some other vma has it mlocked.
146 static void __munlock_isolation_failed(struct page *page)
148 if (PageUnevictable(page))
149 __count_vm_event(UNEVICTABLE_PGSTRANDED);
151 __count_vm_event(UNEVICTABLE_PGMUNLOCKED);
155 * munlock_vma_page - munlock a vma page
156 * @page - page to be unlocked, either a normal page or THP page head
158 * returns the size of the page as a page mask (0 for normal page,
159 * HPAGE_PMD_NR - 1 for THP head page)
161 * called from munlock()/munmap() path with page supposedly on the LRU.
162 * When we munlock a page, because the vma where we found the page is being
163 * munlock()ed or munmap()ed, we want to check whether other vmas hold the
164 * page locked so that we can leave it on the unevictable lru list and not
165 * bother vmscan with it. However, to walk the page's rmap list in
166 * try_to_munlock() we must isolate the page from the LRU. If some other
167 * task has removed the page from the LRU, we won't be able to do that.
168 * So we clear the PageMlocked as we might not get another chance. If we
169 * can't isolate the page, we leave it for putback_lru_page() and vmscan
170 * [page_referenced()/try_to_unmap()] to deal with.
172 unsigned int munlock_vma_page(struct page *page)
174 unsigned int nr_pages;
175 struct zone *zone = page_zone(page);
177 BUG_ON(!PageLocked(page));
180 * Serialize with any parallel __split_huge_page_refcount() which
181 * might otherwise copy PageMlocked to part of the tail pages before
182 * we clear it in the head page. It also stabilizes hpage_nr_pages().
184 spin_lock_irq(&zone->lru_lock);
186 nr_pages = hpage_nr_pages(page);
187 if (!TestClearPageMlocked(page))
190 __mod_zone_page_state(zone, NR_MLOCK, -nr_pages);
192 if (__munlock_isolate_lru_page(page, true)) {
193 spin_unlock_irq(&zone->lru_lock);
194 __munlock_isolated_page(page);
197 __munlock_isolation_failed(page);
200 spin_unlock_irq(&zone->lru_lock);
207 * __mlock_vma_pages_range() - mlock a range of pages in the vma.
209 * @start: start address
212 * This takes care of making the pages present too.
214 * return 0 on success, negative error code on error.
216 * vma->vm_mm->mmap_sem must be held for at least read.
218 long __mlock_vma_pages_range(struct vm_area_struct *vma,
219 unsigned long start, unsigned long end, int *nonblocking)
221 struct mm_struct *mm = vma->vm_mm;
222 unsigned long nr_pages = (end - start) / PAGE_SIZE;
225 VM_BUG_ON(start & ~PAGE_MASK);
226 VM_BUG_ON(end & ~PAGE_MASK);
227 VM_BUG_ON(start < vma->vm_start);
228 VM_BUG_ON(end > vma->vm_end);
229 VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem));
231 gup_flags = FOLL_TOUCH | FOLL_MLOCK;
233 * We want to touch writable mappings with a write fault in order
234 * to break COW, except for shared mappings because these don't COW
235 * and we would not want to dirty them for nothing.
237 if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
238 gup_flags |= FOLL_WRITE;
241 * We want mlock to succeed for regions that have any permissions
242 * other than PROT_NONE.
244 if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))
245 gup_flags |= FOLL_FORCE;
248 * We made sure addr is within a VMA, so the following will
249 * not result in a stack expansion that recurses back here.
251 return __get_user_pages(current, mm, start, nr_pages, gup_flags,
252 NULL, NULL, nonblocking);
256 * convert get_user_pages() return value to posix mlock() error
258 static int __mlock_posix_error_return(long retval)
260 if (retval == -EFAULT)
262 else if (retval == -ENOMEM)
268 * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec()
270 * The fast path is available only for evictable pages with single mapping.
271 * Then we can bypass the per-cpu pvec and get better performance.
272 * when mapcount > 1 we need try_to_munlock() which can fail.
273 * when !page_evictable(), we need the full redo logic of putback_lru_page to
274 * avoid leaving evictable page in unevictable list.
276 * In case of success, @page is added to @pvec and @pgrescued is incremented
277 * in case that the page was previously unevictable. @page is also unlocked.
279 static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec,
282 VM_BUG_ON_PAGE(PageLRU(page), page);
283 VM_BUG_ON_PAGE(!PageLocked(page), page);
285 if (page_mapcount(page) <= 1 && page_evictable(page)) {
286 pagevec_add(pvec, page);
287 if (TestClearPageUnevictable(page))
297 * Putback multiple evictable pages to the LRU
299 * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of
300 * the pages might have meanwhile become unevictable but that is OK.
302 static void __putback_lru_fast(struct pagevec *pvec, int pgrescued)
304 count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec));
306 *__pagevec_lru_add() calls release_pages() so we don't call
307 * put_page() explicitly
309 __pagevec_lru_add(pvec);
310 count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued);
314 * Munlock a batch of pages from the same zone
316 * The work is split to two main phases. First phase clears the Mlocked flag
317 * and attempts to isolate the pages, all under a single zone lru lock.
318 * The second phase finishes the munlock only for pages where isolation
321 * Note that the pagevec may be modified during the process.
323 static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone)
326 int nr = pagevec_count(pvec);
328 struct pagevec pvec_putback;
331 pagevec_init(&pvec_putback, 0);
333 /* Phase 1: page isolation */
334 spin_lock_irq(&zone->lru_lock);
335 for (i = 0; i < nr; i++) {
336 struct page *page = pvec->pages[i];
338 if (TestClearPageMlocked(page)) {
340 * We already have pin from follow_page_mask()
341 * so we can spare the get_page() here.
343 if (__munlock_isolate_lru_page(page, false))
346 __munlock_isolation_failed(page);
350 * We won't be munlocking this page in the next phase
351 * but we still need to release the follow_page_mask()
352 * pin. We cannot do it under lru_lock however. If it's
353 * the last pin, __page_cache_release() would deadlock.
355 pagevec_add(&pvec_putback, pvec->pages[i]);
356 pvec->pages[i] = NULL;
358 delta_munlocked = -nr + pagevec_count(&pvec_putback);
359 __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked);
360 spin_unlock_irq(&zone->lru_lock);
362 /* Now we can release pins of pages that we are not munlocking */
363 pagevec_release(&pvec_putback);
365 /* Phase 2: page munlock */
366 for (i = 0; i < nr; i++) {
367 struct page *page = pvec->pages[i];
371 if (!__putback_lru_fast_prepare(page, &pvec_putback,
374 * Slow path. We don't want to lose the last
375 * pin before unlock_page()
377 get_page(page); /* for putback_lru_page() */
378 __munlock_isolated_page(page);
380 put_page(page); /* from follow_page_mask() */
386 * Phase 3: page putback for pages that qualified for the fast path
387 * This will also call put_page() to return pin from follow_page_mask()
389 if (pagevec_count(&pvec_putback))
390 __putback_lru_fast(&pvec_putback, pgrescued);
394 * Fill up pagevec for __munlock_pagevec using pte walk
396 * The function expects that the struct page corresponding to @start address is
397 * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone.
399 * The rest of @pvec is filled by subsequent pages within the same pmd and same
400 * zone, as long as the pte's are present and vm_normal_page() succeeds. These
401 * pages also get pinned.
403 * Returns the address of the next page that should be scanned. This equals
404 * @start + PAGE_SIZE when no page could be added by the pte walk.
406 static unsigned long __munlock_pagevec_fill(struct pagevec *pvec,
407 struct vm_area_struct *vma, int zoneid, unsigned long start,
414 * Initialize pte walk starting at the already pinned page where we
415 * are sure that there is a pte, as it was pinned under the same
418 pte = get_locked_pte(vma->vm_mm, start, &ptl);
419 /* Make sure we do not cross the page table boundary */
420 end = pgd_addr_end(start, end);
421 end = pud_addr_end(start, end);
422 end = pmd_addr_end(start, end);
424 /* The page next to the pinned page is the first we will try to get */
426 while (start < end) {
427 struct page *page = NULL;
429 if (pte_present(*pte))
430 page = vm_normal_page(vma, start, *pte);
432 * Break if page could not be obtained or the page's node+zone does not
435 if (!page || page_zone_id(page) != zoneid)
440 * Increase the address that will be returned *before* the
441 * eventual break due to pvec becoming full by adding the page
444 if (pagevec_add(pvec, page) == 0)
447 pte_unmap_unlock(pte, ptl);
452 * munlock_vma_pages_range() - munlock all pages in the vma range.'
453 * @vma - vma containing range to be munlock()ed.
454 * @start - start address in @vma of the range
455 * @end - end of range in @vma.
457 * For mremap(), munmap() and exit().
459 * Called with @vma VM_LOCKED.
461 * Returns with VM_LOCKED cleared. Callers must be prepared to
464 * We don't save and restore VM_LOCKED here because pages are
465 * still on lru. In unmap path, pages might be scanned by reclaim
466 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
467 * free them. This will result in freeing mlocked pages.
469 void munlock_vma_pages_range(struct vm_area_struct *vma,
470 unsigned long start, unsigned long end)
472 vma->vm_flags &= ~VM_LOCKED;
474 while (start < end) {
475 struct page *page = NULL;
476 unsigned int page_mask;
477 unsigned long page_increm;
482 pagevec_init(&pvec, 0);
484 * Although FOLL_DUMP is intended for get_dump_page(),
485 * it just so happens that its special treatment of the
486 * ZERO_PAGE (returning an error instead of doing get_page)
487 * suits munlock very well (and if somehow an abnormal page
488 * has sneaked into the range, we won't oops here: great).
490 page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP,
493 if (page && !IS_ERR(page)) {
494 if (PageTransHuge(page)) {
497 * Any THP page found by follow_page_mask() may
498 * have gotten split before reaching
499 * munlock_vma_page(), so we need to recompute
500 * the page_mask here.
502 page_mask = munlock_vma_page(page);
504 put_page(page); /* follow_page_mask() */
507 * Non-huge pages are handled in batches via
508 * pagevec. The pin from follow_page_mask()
509 * prevents them from collapsing by THP.
511 pagevec_add(&pvec, page);
512 zone = page_zone(page);
513 zoneid = page_zone_id(page);
516 * Try to fill the rest of pagevec using fast
517 * pte walk. This will also update start to
518 * the next page to process. Then munlock the
521 start = __munlock_pagevec_fill(&pvec, vma,
523 __munlock_pagevec(&pvec, zone);
527 /* It's a bug to munlock in the middle of a THP page */
528 VM_BUG_ON((start >> PAGE_SHIFT) & page_mask);
529 page_increm = 1 + page_mask;
530 start += page_increm * PAGE_SIZE;
537 * mlock_fixup - handle mlock[all]/munlock[all] requests.
539 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
540 * munlock is a no-op. However, for some special vmas, we go ahead and
543 * For vmas that pass the filters, merge/split as appropriate.
545 static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
546 unsigned long start, unsigned long end, vm_flags_t newflags)
548 struct mm_struct *mm = vma->vm_mm;
552 int lock = !!(newflags & VM_LOCKED);
554 if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
555 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
556 goto out; /* don't set VM_LOCKED, don't count */
558 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
559 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
560 vma->vm_file, pgoff, vma_policy(vma));
566 if (start != vma->vm_start) {
567 ret = split_vma(mm, vma, start, 1);
572 if (end != vma->vm_end) {
573 ret = split_vma(mm, vma, end, 0);
580 * Keep track of amount of locked VM.
582 nr_pages = (end - start) >> PAGE_SHIFT;
584 nr_pages = -nr_pages;
585 mm->locked_vm += nr_pages;
588 * vm_flags is protected by the mmap_sem held in write mode.
589 * It's okay if try_to_unmap_one unmaps a page just after we
590 * set VM_LOCKED, __mlock_vma_pages_range will bring it back.
594 vma->vm_flags = newflags;
596 munlock_vma_pages_range(vma, start, end);
603 static int do_mlock(unsigned long start, size_t len, int on)
605 unsigned long nstart, end, tmp;
606 struct vm_area_struct * vma, * prev;
609 VM_BUG_ON(start & ~PAGE_MASK);
610 VM_BUG_ON(len != PAGE_ALIGN(len));
616 vma = find_vma(current->mm, start);
617 if (!vma || vma->vm_start > start)
621 if (start > vma->vm_start)
624 for (nstart = start ; ; ) {
627 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
629 newflags = vma->vm_flags & ~VM_LOCKED;
631 newflags |= VM_LOCKED;
636 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
640 if (nstart < prev->vm_end)
641 nstart = prev->vm_end;
646 if (!vma || vma->vm_start != nstart) {
655 * __mm_populate - populate and/or mlock pages within a range of address space.
657 * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap
658 * flags. VMAs must be already marked with the desired vm_flags, and
659 * mmap_sem must not be held.
661 int __mm_populate(unsigned long start, unsigned long len, int ignore_errors)
663 struct mm_struct *mm = current->mm;
664 unsigned long end, nstart, nend;
665 struct vm_area_struct *vma = NULL;
669 VM_BUG_ON(start & ~PAGE_MASK);
670 VM_BUG_ON(len != PAGE_ALIGN(len));
673 for (nstart = start; nstart < end; nstart = nend) {
675 * We want to fault in pages for [nstart; end) address range.
676 * Find first corresponding VMA.
680 down_read(&mm->mmap_sem);
681 vma = find_vma(mm, nstart);
682 } else if (nstart >= vma->vm_end)
684 if (!vma || vma->vm_start >= end)
687 * Set [nstart; nend) to intersection of desired address
688 * range with the first VMA. Also, skip undesirable VMA types.
690 nend = min(end, vma->vm_end);
691 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
693 if (nstart < vma->vm_start)
694 nstart = vma->vm_start;
696 * Now fault in a range of pages. __mlock_vma_pages_range()
697 * double checks the vma flags, so that it won't mlock pages
698 * if the vma was already munlocked.
700 ret = __mlock_vma_pages_range(vma, nstart, nend, &locked);
704 continue; /* continue at next VMA */
706 ret = __mlock_posix_error_return(ret);
709 nend = nstart + ret * PAGE_SIZE;
713 up_read(&mm->mmap_sem);
714 return ret; /* 0 or negative error code */
717 SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
719 unsigned long locked;
720 unsigned long lock_limit;
726 lru_add_drain_all(); /* flush pagevec */
728 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
731 lock_limit = rlimit(RLIMIT_MEMLOCK);
732 lock_limit >>= PAGE_SHIFT;
733 locked = len >> PAGE_SHIFT;
735 down_write(¤t->mm->mmap_sem);
737 locked += current->mm->locked_vm;
739 /* check against resource limits */
740 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
741 error = do_mlock(start, len, 1);
743 up_write(¤t->mm->mmap_sem);
745 error = __mm_populate(start, len, 0);
749 SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
753 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
756 down_write(¤t->mm->mmap_sem);
757 ret = do_mlock(start, len, 0);
758 up_write(¤t->mm->mmap_sem);
763 static int do_mlockall(int flags)
765 struct vm_area_struct * vma, * prev = NULL;
767 if (flags & MCL_FUTURE)
768 current->mm->def_flags |= VM_LOCKED;
770 current->mm->def_flags &= ~VM_LOCKED;
771 if (flags == MCL_FUTURE)
774 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
777 newflags = vma->vm_flags & ~VM_LOCKED;
778 if (flags & MCL_CURRENT)
779 newflags |= VM_LOCKED;
782 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
789 SYSCALL_DEFINE1(mlockall, int, flags)
791 unsigned long lock_limit;
794 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
801 if (flags & MCL_CURRENT)
802 lru_add_drain_all(); /* flush pagevec */
804 lock_limit = rlimit(RLIMIT_MEMLOCK);
805 lock_limit >>= PAGE_SHIFT;
808 down_write(¤t->mm->mmap_sem);
810 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
811 capable(CAP_IPC_LOCK))
812 ret = do_mlockall(flags);
813 up_write(¤t->mm->mmap_sem);
814 if (!ret && (flags & MCL_CURRENT))
815 mm_populate(0, TASK_SIZE);
820 SYSCALL_DEFINE0(munlockall)
824 down_write(¤t->mm->mmap_sem);
825 ret = do_mlockall(0);
826 up_write(¤t->mm->mmap_sem);
831 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
832 * shm segments) get accounted against the user_struct instead.
834 static DEFINE_SPINLOCK(shmlock_user_lock);
836 int user_shm_lock(size_t size, struct user_struct *user)
838 unsigned long lock_limit, locked;
841 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
842 lock_limit = rlimit(RLIMIT_MEMLOCK);
843 if (lock_limit == RLIM_INFINITY)
845 lock_limit >>= PAGE_SHIFT;
846 spin_lock(&shmlock_user_lock);
848 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
851 user->locked_shm += locked;
854 spin_unlock(&shmlock_user_lock);
858 void user_shm_unlock(size_t size, struct user_struct *user)
860 spin_lock(&shmlock_user_lock);
861 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
862 spin_unlock(&shmlock_user_lock);