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_zone(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);
193 nr_pages = hpage_nr_pages(page);
194 if (!TestClearPageMlocked(page))
197 __mod_zone_page_state(zone, NR_MLOCK, -nr_pages);
199 if (__munlock_isolate_lru_page(page, true)) {
200 spin_unlock_irq(&zone->lru_lock);
201 __munlock_isolated_page(page);
204 __munlock_isolation_failed(page);
207 spin_unlock_irq(&zone->lru_lock);
214 * convert get_user_pages() return value to posix mlock() error
216 static int __mlock_posix_error_return(long retval)
218 if (retval == -EFAULT)
220 else if (retval == -ENOMEM)
226 * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec()
228 * The fast path is available only for evictable pages with single mapping.
229 * Then we can bypass the per-cpu pvec and get better performance.
230 * when mapcount > 1 we need try_to_munlock() which can fail.
231 * when !page_evictable(), we need the full redo logic of putback_lru_page to
232 * avoid leaving evictable page in unevictable list.
234 * In case of success, @page is added to @pvec and @pgrescued is incremented
235 * in case that the page was previously unevictable. @page is also unlocked.
237 static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec,
240 VM_BUG_ON_PAGE(PageLRU(page), page);
241 VM_BUG_ON_PAGE(!PageLocked(page), page);
243 if (page_mapcount(page) <= 1 && page_evictable(page)) {
244 pagevec_add(pvec, page);
245 if (TestClearPageUnevictable(page))
255 * Putback multiple evictable pages to the LRU
257 * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of
258 * the pages might have meanwhile become unevictable but that is OK.
260 static void __putback_lru_fast(struct pagevec *pvec, int pgrescued)
262 count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec));
264 *__pagevec_lru_add() calls release_pages() so we don't call
265 * put_page() explicitly
267 __pagevec_lru_add(pvec);
268 count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued);
272 * Munlock a batch of pages from the same zone
274 * The work is split to two main phases. First phase clears the Mlocked flag
275 * and attempts to isolate the pages, all under a single zone lru lock.
276 * The second phase finishes the munlock only for pages where isolation
279 * Note that the pagevec may be modified during the process.
281 static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone)
284 int nr = pagevec_count(pvec);
286 struct pagevec pvec_putback;
289 pagevec_init(&pvec_putback, 0);
291 /* Phase 1: page isolation */
292 spin_lock_irq(&zone->lru_lock);
293 for (i = 0; i < nr; i++) {
294 struct page *page = pvec->pages[i];
296 if (TestClearPageMlocked(page)) {
298 * We already have pin from follow_page_mask()
299 * so we can spare the get_page() here.
301 if (__munlock_isolate_lru_page(page, false))
304 __munlock_isolation_failed(page);
308 * We won't be munlocking this page in the next phase
309 * but we still need to release the follow_page_mask()
310 * pin. We cannot do it under lru_lock however. If it's
311 * the last pin, __page_cache_release() would deadlock.
313 pagevec_add(&pvec_putback, pvec->pages[i]);
314 pvec->pages[i] = NULL;
316 delta_munlocked = -nr + pagevec_count(&pvec_putback);
317 __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked);
318 spin_unlock_irq(&zone->lru_lock);
320 /* Now we can release pins of pages that we are not munlocking */
321 pagevec_release(&pvec_putback);
323 /* Phase 2: page munlock */
324 for (i = 0; i < nr; i++) {
325 struct page *page = pvec->pages[i];
329 if (!__putback_lru_fast_prepare(page, &pvec_putback,
332 * Slow path. We don't want to lose the last
333 * pin before unlock_page()
335 get_page(page); /* for putback_lru_page() */
336 __munlock_isolated_page(page);
338 put_page(page); /* from follow_page_mask() */
344 * Phase 3: page putback for pages that qualified for the fast path
345 * This will also call put_page() to return pin from follow_page_mask()
347 if (pagevec_count(&pvec_putback))
348 __putback_lru_fast(&pvec_putback, pgrescued);
352 * Fill up pagevec for __munlock_pagevec using pte walk
354 * The function expects that the struct page corresponding to @start address is
355 * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone.
357 * The rest of @pvec is filled by subsequent pages within the same pmd and same
358 * zone, as long as the pte's are present and vm_normal_page() succeeds. These
359 * pages also get pinned.
361 * Returns the address of the next page that should be scanned. This equals
362 * @start + PAGE_SIZE when no page could be added by the pte walk.
364 static unsigned long __munlock_pagevec_fill(struct pagevec *pvec,
365 struct vm_area_struct *vma, int zoneid, unsigned long start,
372 * Initialize pte walk starting at the already pinned page where we
373 * are sure that there is a pte, as it was pinned under the same
376 pte = get_locked_pte(vma->vm_mm, start, &ptl);
377 /* Make sure we do not cross the page table boundary */
378 end = pgd_addr_end(start, end);
379 end = pud_addr_end(start, end);
380 end = pmd_addr_end(start, end);
382 /* The page next to the pinned page is the first we will try to get */
384 while (start < end) {
385 struct page *page = NULL;
387 if (pte_present(*pte))
388 page = vm_normal_page(vma, start, *pte);
390 * Break if page could not be obtained or the page's node+zone does not
393 if (!page || page_zone_id(page) != zoneid)
397 * Do not use pagevec for PTE-mapped THP,
398 * munlock_vma_pages_range() will handle them.
400 if (PageTransCompound(page))
405 * Increase the address that will be returned *before* the
406 * eventual break due to pvec becoming full by adding the page
409 if (pagevec_add(pvec, page) == 0)
412 pte_unmap_unlock(pte, ptl);
417 * munlock_vma_pages_range() - munlock all pages in the vma range.'
418 * @vma - vma containing range to be munlock()ed.
419 * @start - start address in @vma of the range
420 * @end - end of range in @vma.
422 * For mremap(), munmap() and exit().
424 * Called with @vma VM_LOCKED.
426 * Returns with VM_LOCKED cleared. Callers must be prepared to
429 * We don't save and restore VM_LOCKED here because pages are
430 * still on lru. In unmap path, pages might be scanned by reclaim
431 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
432 * free them. This will result in freeing mlocked pages.
434 void munlock_vma_pages_range(struct vm_area_struct *vma,
435 unsigned long start, unsigned long end)
437 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
439 while (start < end) {
441 unsigned int page_mask;
442 unsigned long page_increm;
447 pagevec_init(&pvec, 0);
449 * Although FOLL_DUMP is intended for get_dump_page(),
450 * it just so happens that its special treatment of the
451 * ZERO_PAGE (returning an error instead of doing get_page)
452 * suits munlock very well (and if somehow an abnormal page
453 * has sneaked into the range, we won't oops here: great).
455 page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP,
458 if (page && !IS_ERR(page)) {
459 if (PageTransTail(page)) {
460 VM_BUG_ON_PAGE(PageMlocked(page), page);
461 put_page(page); /* follow_page_mask() */
462 } else if (PageTransHuge(page)) {
465 * Any THP page found by follow_page_mask() may
466 * have gotten split before reaching
467 * munlock_vma_page(), so we need to recompute
468 * the page_mask here.
470 page_mask = munlock_vma_page(page);
472 put_page(page); /* follow_page_mask() */
475 * Non-huge pages are handled in batches via
476 * pagevec. The pin from follow_page_mask()
477 * prevents them from collapsing by THP.
479 pagevec_add(&pvec, page);
480 zone = page_zone(page);
481 zoneid = page_zone_id(page);
484 * Try to fill the rest of pagevec using fast
485 * pte walk. This will also update start to
486 * the next page to process. Then munlock the
489 start = __munlock_pagevec_fill(&pvec, vma,
491 __munlock_pagevec(&pvec, zone);
495 page_increm = 1 + page_mask;
496 start += page_increm * PAGE_SIZE;
503 * mlock_fixup - handle mlock[all]/munlock[all] requests.
505 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
506 * munlock is a no-op. However, for some special vmas, we go ahead and
509 * For vmas that pass the filters, merge/split as appropriate.
511 static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
512 unsigned long start, unsigned long end, vm_flags_t newflags)
514 struct mm_struct *mm = vma->vm_mm;
518 int lock = !!(newflags & VM_LOCKED);
520 if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
521 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
522 /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
525 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
526 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
527 vma->vm_file, pgoff, vma_policy(vma),
528 vma->vm_userfaultfd_ctx);
534 if (start != vma->vm_start) {
535 ret = split_vma(mm, vma, start, 1);
540 if (end != vma->vm_end) {
541 ret = split_vma(mm, vma, end, 0);
548 * Keep track of amount of locked VM.
550 nr_pages = (end - start) >> PAGE_SHIFT;
552 nr_pages = -nr_pages;
553 mm->locked_vm += nr_pages;
556 * vm_flags is protected by the mmap_sem held in write mode.
557 * It's okay if try_to_unmap_one unmaps a page just after we
558 * set VM_LOCKED, populate_vma_page_range will bring it back.
562 vma->vm_flags = newflags;
564 munlock_vma_pages_range(vma, start, end);
571 static int apply_vma_lock_flags(unsigned long start, size_t len,
574 unsigned long nstart, end, tmp;
575 struct vm_area_struct * vma, * prev;
578 VM_BUG_ON(offset_in_page(start));
579 VM_BUG_ON(len != PAGE_ALIGN(len));
585 vma = find_vma(current->mm, start);
586 if (!vma || vma->vm_start > start)
590 if (start > vma->vm_start)
593 for (nstart = start ; ; ) {
594 vm_flags_t newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
598 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
602 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
606 if (nstart < prev->vm_end)
607 nstart = prev->vm_end;
612 if (!vma || vma->vm_start != nstart) {
620 static int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
622 unsigned long locked;
623 unsigned long lock_limit;
629 lru_add_drain_all(); /* flush pagevec */
631 len = PAGE_ALIGN(len + (offset_in_page(start)));
634 lock_limit = rlimit(RLIMIT_MEMLOCK);
635 lock_limit >>= PAGE_SHIFT;
636 locked = len >> PAGE_SHIFT;
638 down_write(¤t->mm->mmap_sem);
640 locked += current->mm->locked_vm;
642 /* check against resource limits */
643 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
644 error = apply_vma_lock_flags(start, len, flags);
646 up_write(¤t->mm->mmap_sem);
650 error = __mm_populate(start, len, 0);
652 return __mlock_posix_error_return(error);
656 SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
658 return do_mlock(start, len, VM_LOCKED);
661 SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
663 vm_flags_t vm_flags = VM_LOCKED;
665 if (flags & ~MLOCK_ONFAULT)
668 if (flags & MLOCK_ONFAULT)
669 vm_flags |= VM_LOCKONFAULT;
671 return do_mlock(start, len, vm_flags);
674 SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
678 len = PAGE_ALIGN(len + (offset_in_page(start)));
681 down_write(¤t->mm->mmap_sem);
682 ret = apply_vma_lock_flags(start, len, 0);
683 up_write(¤t->mm->mmap_sem);
689 * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
690 * and translate into the appropriate modifications to mm->def_flags and/or the
691 * flags for all current VMAs.
693 * There are a couple of subtleties with this. If mlockall() is called multiple
694 * times with different flags, the values do not necessarily stack. If mlockall
695 * is called once including the MCL_FUTURE flag and then a second time without
696 * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
698 static int apply_mlockall_flags(int flags)
700 struct vm_area_struct * vma, * prev = NULL;
701 vm_flags_t to_add = 0;
703 current->mm->def_flags &= VM_LOCKED_CLEAR_MASK;
704 if (flags & MCL_FUTURE) {
705 current->mm->def_flags |= VM_LOCKED;
707 if (flags & MCL_ONFAULT)
708 current->mm->def_flags |= VM_LOCKONFAULT;
710 if (!(flags & MCL_CURRENT))
714 if (flags & MCL_CURRENT) {
716 if (flags & MCL_ONFAULT)
717 to_add |= VM_LOCKONFAULT;
720 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
723 newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
727 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
728 cond_resched_rcu_qs();
734 SYSCALL_DEFINE1(mlockall, int, flags)
736 unsigned long lock_limit;
739 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)))
745 if (flags & MCL_CURRENT)
746 lru_add_drain_all(); /* flush pagevec */
748 lock_limit = rlimit(RLIMIT_MEMLOCK);
749 lock_limit >>= PAGE_SHIFT;
752 down_write(¤t->mm->mmap_sem);
754 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
755 capable(CAP_IPC_LOCK))
756 ret = apply_mlockall_flags(flags);
757 up_write(¤t->mm->mmap_sem);
758 if (!ret && (flags & MCL_CURRENT))
759 mm_populate(0, TASK_SIZE);
764 SYSCALL_DEFINE0(munlockall)
768 down_write(¤t->mm->mmap_sem);
769 ret = apply_mlockall_flags(0);
770 up_write(¤t->mm->mmap_sem);
775 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
776 * shm segments) get accounted against the user_struct instead.
778 static DEFINE_SPINLOCK(shmlock_user_lock);
780 int user_shm_lock(size_t size, struct user_struct *user)
782 unsigned long lock_limit, locked;
785 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
786 lock_limit = rlimit(RLIMIT_MEMLOCK);
787 if (lock_limit == RLIM_INFINITY)
789 lock_limit >>= PAGE_SHIFT;
790 spin_lock(&shmlock_user_lock);
792 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
795 user->locked_shm += locked;
798 spin_unlock(&shmlock_user_lock);
802 void user_shm_unlock(size_t size, struct user_struct *user)
804 spin_lock(&shmlock_user_lock);
805 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
806 spin_unlock(&shmlock_user_lock);