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 bool can_do_mlock(void)
29 if (rlimit(RLIMIT_MEMLOCK) != 0)
31 if (capable(CAP_IPC_LOCK))
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 /* Serialize with page migration */
83 BUG_ON(!PageLocked(page));
85 VM_BUG_ON_PAGE(PageTail(page), page);
86 VM_BUG_ON_PAGE(PageCompound(page) && PageDoubleMap(page), page);
88 if (!TestSetPageMlocked(page)) {
89 mod_zone_page_state(page_zone(page), NR_MLOCK,
90 hpage_nr_pages(page));
91 count_vm_event(UNEVICTABLE_PGMLOCKED);
92 if (!isolate_lru_page(page))
93 putback_lru_page(page);
98 * Isolate a page from LRU with optional get_page() pin.
99 * Assumes lru_lock already held and page already pinned.
101 static bool __munlock_isolate_lru_page(struct page *page, bool getpage)
104 struct lruvec *lruvec;
106 lruvec = mem_cgroup_page_lruvec(page, page_pgdat(page));
110 del_page_from_lru_list(page, lruvec, page_lru(page));
118 * Finish munlock after successful page isolation
120 * Page must be locked. This is a wrapper for try_to_munlock()
121 * and putback_lru_page() with munlock accounting.
123 static void __munlock_isolated_page(struct page *page)
125 int ret = SWAP_AGAIN;
128 * Optimization: if the page was mapped just once, that's our mapping
129 * and we don't need to check all the other vmas.
131 if (page_mapcount(page) > 1)
132 ret = try_to_munlock(page);
134 /* Did try_to_unlock() succeed or punt? */
135 if (ret != SWAP_MLOCK)
136 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
138 putback_lru_page(page);
142 * Accounting for page isolation fail during munlock
144 * Performs accounting when page isolation fails in munlock. There is nothing
145 * else to do because it means some other task has already removed the page
146 * from the LRU. putback_lru_page() will take care of removing the page from
147 * the unevictable list, if necessary. vmscan [page_referenced()] will move
148 * the page back to the unevictable list if some other vma has it mlocked.
150 static void __munlock_isolation_failed(struct page *page)
152 if (PageUnevictable(page))
153 __count_vm_event(UNEVICTABLE_PGSTRANDED);
155 __count_vm_event(UNEVICTABLE_PGMUNLOCKED);
159 * munlock_vma_page - munlock a vma page
160 * @page - page to be unlocked, either a normal page or THP page head
162 * returns the size of the page as a page mask (0 for normal page,
163 * HPAGE_PMD_NR - 1 for THP head page)
165 * called from munlock()/munmap() path with page supposedly on the LRU.
166 * When we munlock a page, because the vma where we found the page is being
167 * munlock()ed or munmap()ed, we want to check whether other vmas hold the
168 * page locked so that we can leave it on the unevictable lru list and not
169 * bother vmscan with it. However, to walk the page's rmap list in
170 * try_to_munlock() we must isolate the page from the LRU. If some other
171 * task has removed the page from the LRU, we won't be able to do that.
172 * So we clear the PageMlocked as we might not get another chance. If we
173 * can't isolate the page, we leave it for putback_lru_page() and vmscan
174 * [page_referenced()/try_to_unmap()] to deal with.
176 unsigned int munlock_vma_page(struct page *page)
179 struct zone *zone = page_zone(page);
181 /* For try_to_munlock() and to serialize with page migration */
182 BUG_ON(!PageLocked(page));
184 VM_BUG_ON_PAGE(PageTail(page), page);
187 * Serialize with any parallel __split_huge_page_refcount() which
188 * might otherwise copy PageMlocked to part of the tail pages before
189 * we clear it in the head page. It also stabilizes hpage_nr_pages().
191 spin_lock_irq(zone_lru_lock(zone));
193 if (!TestClearPageMlocked(page)) {
194 /* Potentially, PTE-mapped THP: do not skip the rest PTEs */
199 nr_pages = hpage_nr_pages(page);
200 __mod_zone_page_state(zone, NR_MLOCK, -nr_pages);
202 if (__munlock_isolate_lru_page(page, true)) {
203 spin_unlock_irq(zone_lru_lock(zone));
204 __munlock_isolated_page(page);
207 __munlock_isolation_failed(page);
210 spin_unlock_irq(zone_lru_lock(zone));
217 * convert get_user_pages() return value to posix mlock() error
219 static int __mlock_posix_error_return(long retval)
221 if (retval == -EFAULT)
223 else if (retval == -ENOMEM)
229 * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec()
231 * The fast path is available only for evictable pages with single mapping.
232 * Then we can bypass the per-cpu pvec and get better performance.
233 * when mapcount > 1 we need try_to_munlock() which can fail.
234 * when !page_evictable(), we need the full redo logic of putback_lru_page to
235 * avoid leaving evictable page in unevictable list.
237 * In case of success, @page is added to @pvec and @pgrescued is incremented
238 * in case that the page was previously unevictable. @page is also unlocked.
240 static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec,
243 VM_BUG_ON_PAGE(PageLRU(page), page);
244 VM_BUG_ON_PAGE(!PageLocked(page), page);
246 if (page_mapcount(page) <= 1 && page_evictable(page)) {
247 pagevec_add(pvec, page);
248 if (TestClearPageUnevictable(page))
258 * Putback multiple evictable pages to the LRU
260 * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of
261 * the pages might have meanwhile become unevictable but that is OK.
263 static void __putback_lru_fast(struct pagevec *pvec, int pgrescued)
265 count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec));
267 *__pagevec_lru_add() calls release_pages() so we don't call
268 * put_page() explicitly
270 __pagevec_lru_add(pvec);
271 count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued);
275 * Munlock a batch of pages from the same zone
277 * The work is split to two main phases. First phase clears the Mlocked flag
278 * and attempts to isolate the pages, all under a single zone lru lock.
279 * The second phase finishes the munlock only for pages where isolation
282 * Note that the pagevec may be modified during the process.
284 static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone)
287 int nr = pagevec_count(pvec);
289 struct pagevec pvec_putback;
292 pagevec_init(&pvec_putback, 0);
294 /* Phase 1: page isolation */
295 spin_lock_irq(zone_lru_lock(zone));
296 for (i = 0; i < nr; i++) {
297 struct page *page = pvec->pages[i];
299 if (TestClearPageMlocked(page)) {
301 * We already have pin from follow_page_mask()
302 * so we can spare the get_page() here.
304 if (__munlock_isolate_lru_page(page, false))
307 __munlock_isolation_failed(page);
311 * We won't be munlocking this page in the next phase
312 * but we still need to release the follow_page_mask()
313 * pin. We cannot do it under lru_lock however. If it's
314 * the last pin, __page_cache_release() would deadlock.
316 pagevec_add(&pvec_putback, pvec->pages[i]);
317 pvec->pages[i] = NULL;
319 delta_munlocked = -nr + pagevec_count(&pvec_putback);
320 __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked);
321 spin_unlock_irq(zone_lru_lock(zone));
323 /* Now we can release pins of pages that we are not munlocking */
324 pagevec_release(&pvec_putback);
326 /* Phase 2: page munlock */
327 for (i = 0; i < nr; i++) {
328 struct page *page = pvec->pages[i];
332 if (!__putback_lru_fast_prepare(page, &pvec_putback,
335 * Slow path. We don't want to lose the last
336 * pin before unlock_page()
338 get_page(page); /* for putback_lru_page() */
339 __munlock_isolated_page(page);
341 put_page(page); /* from follow_page_mask() */
347 * Phase 3: page putback for pages that qualified for the fast path
348 * This will also call put_page() to return pin from follow_page_mask()
350 if (pagevec_count(&pvec_putback))
351 __putback_lru_fast(&pvec_putback, pgrescued);
355 * Fill up pagevec for __munlock_pagevec using pte walk
357 * The function expects that the struct page corresponding to @start address is
358 * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone.
360 * The rest of @pvec is filled by subsequent pages within the same pmd and same
361 * zone, as long as the pte's are present and vm_normal_page() succeeds. These
362 * pages also get pinned.
364 * Returns the address of the next page that should be scanned. This equals
365 * @start + PAGE_SIZE when no page could be added by the pte walk.
367 static unsigned long __munlock_pagevec_fill(struct pagevec *pvec,
368 struct vm_area_struct *vma, int zoneid, unsigned long start,
375 * Initialize pte walk starting at the already pinned page where we
376 * are sure that there is a pte, as it was pinned under the same
379 pte = get_locked_pte(vma->vm_mm, start, &ptl);
380 /* Make sure we do not cross the page table boundary */
381 end = pgd_addr_end(start, end);
382 end = pud_addr_end(start, end);
383 end = pmd_addr_end(start, end);
385 /* The page next to the pinned page is the first we will try to get */
387 while (start < end) {
388 struct page *page = NULL;
390 if (pte_present(*pte))
391 page = vm_normal_page(vma, start, *pte);
393 * Break if page could not be obtained or the page's node+zone does not
396 if (!page || page_zone_id(page) != zoneid)
400 * Do not use pagevec for PTE-mapped THP,
401 * munlock_vma_pages_range() will handle them.
403 if (PageTransCompound(page))
408 * Increase the address that will be returned *before* the
409 * eventual break due to pvec becoming full by adding the page
412 if (pagevec_add(pvec, page) == 0)
415 pte_unmap_unlock(pte, ptl);
420 * munlock_vma_pages_range() - munlock all pages in the vma range.'
421 * @vma - vma containing range to be munlock()ed.
422 * @start - start address in @vma of the range
423 * @end - end of range in @vma.
425 * For mremap(), munmap() and exit().
427 * Called with @vma VM_LOCKED.
429 * Returns with VM_LOCKED cleared. Callers must be prepared to
432 * We don't save and restore VM_LOCKED here because pages are
433 * still on lru. In unmap path, pages might be scanned by reclaim
434 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
435 * free them. This will result in freeing mlocked pages.
437 void munlock_vma_pages_range(struct vm_area_struct *vma,
438 unsigned long start, unsigned long end)
440 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
442 while (start < end) {
444 unsigned int page_mask;
445 unsigned long page_increm;
450 pagevec_init(&pvec, 0);
452 * Although FOLL_DUMP is intended for get_dump_page(),
453 * it just so happens that its special treatment of the
454 * ZERO_PAGE (returning an error instead of doing get_page)
455 * suits munlock very well (and if somehow an abnormal page
456 * has sneaked into the range, we won't oops here: great).
458 page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP,
461 if (page && !IS_ERR(page)) {
462 if (PageTransTail(page)) {
463 VM_BUG_ON_PAGE(PageMlocked(page), page);
464 put_page(page); /* follow_page_mask() */
465 } else if (PageTransHuge(page)) {
468 * Any THP page found by follow_page_mask() may
469 * have gotten split before reaching
470 * munlock_vma_page(), so we need to recompute
471 * the page_mask here.
473 page_mask = munlock_vma_page(page);
475 put_page(page); /* follow_page_mask() */
478 * Non-huge pages are handled in batches via
479 * pagevec. The pin from follow_page_mask()
480 * prevents them from collapsing by THP.
482 pagevec_add(&pvec, page);
483 zone = page_zone(page);
484 zoneid = page_zone_id(page);
487 * Try to fill the rest of pagevec using fast
488 * pte walk. This will also update start to
489 * the next page to process. Then munlock the
492 start = __munlock_pagevec_fill(&pvec, vma,
494 __munlock_pagevec(&pvec, zone);
498 page_increm = 1 + page_mask;
499 start += page_increm * PAGE_SIZE;
506 * mlock_fixup - handle mlock[all]/munlock[all] requests.
508 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
509 * munlock is a no-op. However, for some special vmas, we go ahead and
512 * For vmas that pass the filters, merge/split as appropriate.
514 static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
515 unsigned long start, unsigned long end, vm_flags_t newflags)
517 struct mm_struct *mm = vma->vm_mm;
521 int lock = !!(newflags & VM_LOCKED);
522 vm_flags_t old_flags = vma->vm_flags;
524 if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
525 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
526 /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
529 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
530 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
531 vma->vm_file, pgoff, vma_policy(vma),
532 vma->vm_userfaultfd_ctx);
538 if (start != vma->vm_start) {
539 ret = split_vma(mm, vma, start, 1);
544 if (end != vma->vm_end) {
545 ret = split_vma(mm, vma, end, 0);
552 * Keep track of amount of locked VM.
554 nr_pages = (end - start) >> PAGE_SHIFT;
556 nr_pages = -nr_pages;
557 else if (old_flags & VM_LOCKED)
559 mm->locked_vm += nr_pages;
562 * vm_flags is protected by the mmap_sem held in write mode.
563 * It's okay if try_to_unmap_one unmaps a page just after we
564 * set VM_LOCKED, populate_vma_page_range will bring it back.
568 vma->vm_flags = newflags;
570 munlock_vma_pages_range(vma, start, end);
577 static int apply_vma_lock_flags(unsigned long start, size_t len,
580 unsigned long nstart, end, tmp;
581 struct vm_area_struct * vma, * prev;
584 VM_BUG_ON(offset_in_page(start));
585 VM_BUG_ON(len != PAGE_ALIGN(len));
591 vma = find_vma(current->mm, start);
592 if (!vma || vma->vm_start > start)
596 if (start > vma->vm_start)
599 for (nstart = start ; ; ) {
600 vm_flags_t newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
604 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
608 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
612 if (nstart < prev->vm_end)
613 nstart = prev->vm_end;
618 if (!vma || vma->vm_start != nstart) {
627 * Go through vma areas and sum size of mlocked
628 * vma pages, as return value.
629 * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT)
631 * Return value: previously mlocked page counts
633 static int count_mm_mlocked_page_nr(struct mm_struct *mm,
634 unsigned long start, size_t len)
636 struct vm_area_struct *vma;
642 vma = find_vma(mm, start);
646 for (; vma ; vma = vma->vm_next) {
647 if (start >= vma->vm_end)
649 if (start + len <= vma->vm_start)
651 if (vma->vm_flags & VM_LOCKED) {
652 if (start > vma->vm_start)
653 count -= (start - vma->vm_start);
654 if (start + len < vma->vm_end) {
655 count += start + len - vma->vm_start;
658 count += vma->vm_end - vma->vm_start;
662 return count >> PAGE_SHIFT;
665 static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
667 unsigned long locked;
668 unsigned long lock_limit;
674 lru_add_drain_all(); /* flush pagevec */
676 len = PAGE_ALIGN(len + (offset_in_page(start)));
679 lock_limit = rlimit(RLIMIT_MEMLOCK);
680 lock_limit >>= PAGE_SHIFT;
681 locked = len >> PAGE_SHIFT;
683 if (down_write_killable(¤t->mm->mmap_sem))
686 locked += current->mm->locked_vm;
687 if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) {
689 * It is possible that the regions requested intersect with
690 * previously mlocked areas, that part area in "mm->locked_vm"
691 * should not be counted to new mlock increment count. So check
692 * and adjust locked count if necessary.
694 locked -= count_mm_mlocked_page_nr(current->mm,
698 /* check against resource limits */
699 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
700 error = apply_vma_lock_flags(start, len, flags);
702 up_write(¤t->mm->mmap_sem);
706 error = __mm_populate(start, len, 0);
708 return __mlock_posix_error_return(error);
712 SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
714 return do_mlock(start, len, VM_LOCKED);
717 SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
719 vm_flags_t vm_flags = VM_LOCKED;
721 if (flags & ~MLOCK_ONFAULT)
724 if (flags & MLOCK_ONFAULT)
725 vm_flags |= VM_LOCKONFAULT;
727 return do_mlock(start, len, vm_flags);
730 SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
734 len = PAGE_ALIGN(len + (offset_in_page(start)));
737 if (down_write_killable(¤t->mm->mmap_sem))
739 ret = apply_vma_lock_flags(start, len, 0);
740 up_write(¤t->mm->mmap_sem);
746 * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
747 * and translate into the appropriate modifications to mm->def_flags and/or the
748 * flags for all current VMAs.
750 * There are a couple of subtleties with this. If mlockall() is called multiple
751 * times with different flags, the values do not necessarily stack. If mlockall
752 * is called once including the MCL_FUTURE flag and then a second time without
753 * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
755 static int apply_mlockall_flags(int flags)
757 struct vm_area_struct * vma, * prev = NULL;
758 vm_flags_t to_add = 0;
760 current->mm->def_flags &= VM_LOCKED_CLEAR_MASK;
761 if (flags & MCL_FUTURE) {
762 current->mm->def_flags |= VM_LOCKED;
764 if (flags & MCL_ONFAULT)
765 current->mm->def_flags |= VM_LOCKONFAULT;
767 if (!(flags & MCL_CURRENT))
771 if (flags & MCL_CURRENT) {
773 if (flags & MCL_ONFAULT)
774 to_add |= VM_LOCKONFAULT;
777 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
780 newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
784 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
785 cond_resched_rcu_qs();
791 SYSCALL_DEFINE1(mlockall, int, flags)
793 unsigned long lock_limit;
796 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)))
802 if (flags & MCL_CURRENT)
803 lru_add_drain_all(); /* flush pagevec */
805 lock_limit = rlimit(RLIMIT_MEMLOCK);
806 lock_limit >>= PAGE_SHIFT;
808 if (down_write_killable(¤t->mm->mmap_sem))
812 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
813 capable(CAP_IPC_LOCK))
814 ret = apply_mlockall_flags(flags);
815 up_write(¤t->mm->mmap_sem);
816 if (!ret && (flags & MCL_CURRENT))
817 mm_populate(0, TASK_SIZE);
822 SYSCALL_DEFINE0(munlockall)
826 if (down_write_killable(¤t->mm->mmap_sem))
828 ret = apply_mlockall_flags(0);
829 up_write(¤t->mm->mmap_sem);
834 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
835 * shm segments) get accounted against the user_struct instead.
837 static DEFINE_SPINLOCK(shmlock_user_lock);
839 int user_shm_lock(size_t size, struct user_struct *user)
841 unsigned long lock_limit, locked;
844 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
845 lock_limit = rlimit(RLIMIT_MEMLOCK);
846 if (lock_limit == RLIM_INFINITY)
848 lock_limit >>= PAGE_SHIFT;
849 spin_lock(&shmlock_user_lock);
851 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
854 user->locked_shm += locked;
857 spin_unlock(&shmlock_user_lock);
861 void user_shm_unlock(size_t size, struct user_struct *user)
863 spin_lock(&shmlock_user_lock);
864 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
865 spin_unlock(&shmlock_user_lock);