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 if (!TestSetPageMlocked(page)) {
83 mod_zone_page_state(page_zone(page), NR_MLOCK,
84 hpage_nr_pages(page));
85 count_vm_event(UNEVICTABLE_PGMLOCKED);
86 if (!isolate_lru_page(page))
87 putback_lru_page(page);
92 * Isolate a page from LRU with optional get_page() pin.
93 * Assumes lru_lock already held and page already pinned.
95 static bool __munlock_isolate_lru_page(struct page *page, bool getpage)
98 struct lruvec *lruvec;
100 lruvec = mem_cgroup_page_lruvec(page, page_zone(page));
104 del_page_from_lru_list(page, lruvec, page_lru(page));
112 * Finish munlock after successful page isolation
114 * Page must be locked. This is a wrapper for try_to_munlock()
115 * and putback_lru_page() with munlock accounting.
117 static void __munlock_isolated_page(struct page *page)
119 int ret = SWAP_AGAIN;
122 * Optimization: if the page was mapped just once, that's our mapping
123 * and we don't need to check all the other vmas.
125 if (page_mapcount(page) > 1)
126 ret = try_to_munlock(page);
128 /* Did try_to_unlock() succeed or punt? */
129 if (ret != SWAP_MLOCK)
130 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
132 putback_lru_page(page);
136 * Accounting for page isolation fail during munlock
138 * Performs accounting when page isolation fails in munlock. There is nothing
139 * else to do because it means some other task has already removed the page
140 * from the LRU. putback_lru_page() will take care of removing the page from
141 * the unevictable list, if necessary. vmscan [page_referenced()] will move
142 * the page back to the unevictable list if some other vma has it mlocked.
144 static void __munlock_isolation_failed(struct page *page)
146 if (PageUnevictable(page))
147 __count_vm_event(UNEVICTABLE_PGSTRANDED);
149 __count_vm_event(UNEVICTABLE_PGMUNLOCKED);
153 * munlock_vma_page - munlock a vma page
154 * @page - page to be unlocked, either a normal page or THP page head
156 * returns the size of the page as a page mask (0 for normal page,
157 * HPAGE_PMD_NR - 1 for THP head page)
159 * called from munlock()/munmap() path with page supposedly on the LRU.
160 * When we munlock a page, because the vma where we found the page is being
161 * munlock()ed or munmap()ed, we want to check whether other vmas hold the
162 * page locked so that we can leave it on the unevictable lru list and not
163 * bother vmscan with it. However, to walk the page's rmap list in
164 * try_to_munlock() we must isolate the page from the LRU. If some other
165 * task has removed the page from the LRU, we won't be able to do that.
166 * So we clear the PageMlocked as we might not get another chance. If we
167 * can't isolate the page, we leave it for putback_lru_page() and vmscan
168 * [page_referenced()/try_to_unmap()] to deal with.
170 unsigned int munlock_vma_page(struct page *page)
172 unsigned int nr_pages;
173 struct zone *zone = page_zone(page);
175 BUG_ON(!PageLocked(page));
178 * Serialize with any parallel __split_huge_page_refcount() which
179 * might otherwise copy PageMlocked to part of the tail pages before
180 * we clear it in the head page. It also stabilizes hpage_nr_pages().
182 spin_lock_irq(&zone->lru_lock);
184 nr_pages = hpage_nr_pages(page);
185 if (!TestClearPageMlocked(page))
188 __mod_zone_page_state(zone, NR_MLOCK, -nr_pages);
190 if (__munlock_isolate_lru_page(page, true)) {
191 spin_unlock_irq(&zone->lru_lock);
192 __munlock_isolated_page(page);
195 __munlock_isolation_failed(page);
198 spin_unlock_irq(&zone->lru_lock);
205 * __mlock_vma_pages_range() - mlock a range of pages in the vma.
207 * @start: start address
210 * This takes care of making the pages present too.
212 * return 0 on success, negative error code on error.
214 * vma->vm_mm->mmap_sem must be held for at least read.
216 long __mlock_vma_pages_range(struct vm_area_struct *vma,
217 unsigned long start, unsigned long end, int *nonblocking)
219 struct mm_struct *mm = vma->vm_mm;
220 unsigned long nr_pages = (end - start) / PAGE_SIZE;
223 VM_BUG_ON(start & ~PAGE_MASK);
224 VM_BUG_ON(end & ~PAGE_MASK);
225 VM_BUG_ON(start < vma->vm_start);
226 VM_BUG_ON(end > vma->vm_end);
227 VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem));
229 gup_flags = FOLL_TOUCH | FOLL_MLOCK;
231 * We want to touch writable mappings with a write fault in order
232 * to break COW, except for shared mappings because these don't COW
233 * and we would not want to dirty them for nothing.
235 if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
236 gup_flags |= FOLL_WRITE;
239 * We want mlock to succeed for regions that have any permissions
240 * other than PROT_NONE.
242 if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))
243 gup_flags |= FOLL_FORCE;
246 * We made sure addr is within a VMA, so the following will
247 * not result in a stack expansion that recurses back here.
249 return __get_user_pages(current, mm, start, nr_pages, gup_flags,
250 NULL, NULL, nonblocking);
254 * convert get_user_pages() return value to posix mlock() error
256 static int __mlock_posix_error_return(long retval)
258 if (retval == -EFAULT)
260 else if (retval == -ENOMEM)
266 * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec()
268 * The fast path is available only for evictable pages with single mapping.
269 * Then we can bypass the per-cpu pvec and get better performance.
270 * when mapcount > 1 we need try_to_munlock() which can fail.
271 * when !page_evictable(), we need the full redo logic of putback_lru_page to
272 * avoid leaving evictable page in unevictable list.
274 * In case of success, @page is added to @pvec and @pgrescued is incremented
275 * in case that the page was previously unevictable. @page is also unlocked.
277 static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec,
280 VM_BUG_ON_PAGE(PageLRU(page), page);
281 VM_BUG_ON_PAGE(!PageLocked(page), page);
283 if (page_mapcount(page) <= 1 && page_evictable(page)) {
284 pagevec_add(pvec, page);
285 if (TestClearPageUnevictable(page))
295 * Putback multiple evictable pages to the LRU
297 * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of
298 * the pages might have meanwhile become unevictable but that is OK.
300 static void __putback_lru_fast(struct pagevec *pvec, int pgrescued)
302 count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec));
304 *__pagevec_lru_add() calls release_pages() so we don't call
305 * put_page() explicitly
307 __pagevec_lru_add(pvec);
308 count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued);
312 * Munlock a batch of pages from the same zone
314 * The work is split to two main phases. First phase clears the Mlocked flag
315 * and attempts to isolate the pages, all under a single zone lru lock.
316 * The second phase finishes the munlock only for pages where isolation
319 * Note that the pagevec may be modified during the process.
321 static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone)
324 int nr = pagevec_count(pvec);
326 struct pagevec pvec_putback;
329 pagevec_init(&pvec_putback, 0);
331 /* Phase 1: page isolation */
332 spin_lock_irq(&zone->lru_lock);
333 for (i = 0; i < nr; i++) {
334 struct page *page = pvec->pages[i];
336 if (TestClearPageMlocked(page)) {
338 * We already have pin from follow_page_mask()
339 * so we can spare the get_page() here.
341 if (__munlock_isolate_lru_page(page, false))
344 __munlock_isolation_failed(page);
348 * We won't be munlocking this page in the next phase
349 * but we still need to release the follow_page_mask()
350 * pin. We cannot do it under lru_lock however. If it's
351 * the last pin, __page_cache_release() would deadlock.
353 pagevec_add(&pvec_putback, pvec->pages[i]);
354 pvec->pages[i] = NULL;
356 delta_munlocked = -nr + pagevec_count(&pvec_putback);
357 __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked);
358 spin_unlock_irq(&zone->lru_lock);
360 /* Now we can release pins of pages that we are not munlocking */
361 pagevec_release(&pvec_putback);
363 /* Phase 2: page munlock */
364 for (i = 0; i < nr; i++) {
365 struct page *page = pvec->pages[i];
369 if (!__putback_lru_fast_prepare(page, &pvec_putback,
372 * Slow path. We don't want to lose the last
373 * pin before unlock_page()
375 get_page(page); /* for putback_lru_page() */
376 __munlock_isolated_page(page);
378 put_page(page); /* from follow_page_mask() */
384 * Phase 3: page putback for pages that qualified for the fast path
385 * This will also call put_page() to return pin from follow_page_mask()
387 if (pagevec_count(&pvec_putback))
388 __putback_lru_fast(&pvec_putback, pgrescued);
392 * Fill up pagevec for __munlock_pagevec using pte walk
394 * The function expects that the struct page corresponding to @start address is
395 * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone.
397 * The rest of @pvec is filled by subsequent pages within the same pmd and same
398 * zone, as long as the pte's are present and vm_normal_page() succeeds. These
399 * pages also get pinned.
401 * Returns the address of the next page that should be scanned. This equals
402 * @start + PAGE_SIZE when no page could be added by the pte walk.
404 static unsigned long __munlock_pagevec_fill(struct pagevec *pvec,
405 struct vm_area_struct *vma, int zoneid, unsigned long start,
412 * Initialize pte walk starting at the already pinned page where we
413 * are sure that there is a pte, as it was pinned under the same
416 pte = get_locked_pte(vma->vm_mm, start, &ptl);
417 /* Make sure we do not cross the page table boundary */
418 end = pgd_addr_end(start, end);
419 end = pud_addr_end(start, end);
420 end = pmd_addr_end(start, end);
422 /* The page next to the pinned page is the first we will try to get */
424 while (start < end) {
425 struct page *page = NULL;
427 if (pte_present(*pte))
428 page = vm_normal_page(vma, start, *pte);
430 * Break if page could not be obtained or the page's node+zone does not
433 if (!page || page_zone_id(page) != zoneid)
438 * Increase the address that will be returned *before* the
439 * eventual break due to pvec becoming full by adding the page
442 if (pagevec_add(pvec, page) == 0)
445 pte_unmap_unlock(pte, ptl);
450 * munlock_vma_pages_range() - munlock all pages in the vma range.'
451 * @vma - vma containing range to be munlock()ed.
452 * @start - start address in @vma of the range
453 * @end - end of range in @vma.
455 * For mremap(), munmap() and exit().
457 * Called with @vma VM_LOCKED.
459 * Returns with VM_LOCKED cleared. Callers must be prepared to
462 * We don't save and restore VM_LOCKED here because pages are
463 * still on lru. In unmap path, pages might be scanned by reclaim
464 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
465 * free them. This will result in freeing mlocked pages.
467 void munlock_vma_pages_range(struct vm_area_struct *vma,
468 unsigned long start, unsigned long end)
470 vma->vm_flags &= ~VM_LOCKED;
472 while (start < end) {
473 struct page *page = NULL;
474 unsigned int page_mask;
475 unsigned long page_increm;
480 pagevec_init(&pvec, 0);
482 * Although FOLL_DUMP is intended for get_dump_page(),
483 * it just so happens that its special treatment of the
484 * ZERO_PAGE (returning an error instead of doing get_page)
485 * suits munlock very well (and if somehow an abnormal page
486 * has sneaked into the range, we won't oops here: great).
488 page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP,
491 if (page && !IS_ERR(page)) {
492 if (PageTransHuge(page)) {
495 * Any THP page found by follow_page_mask() may
496 * have gotten split before reaching
497 * munlock_vma_page(), so we need to recompute
498 * the page_mask here.
500 page_mask = munlock_vma_page(page);
502 put_page(page); /* follow_page_mask() */
505 * Non-huge pages are handled in batches via
506 * pagevec. The pin from follow_page_mask()
507 * prevents them from collapsing by THP.
509 pagevec_add(&pvec, page);
510 zone = page_zone(page);
511 zoneid = page_zone_id(page);
514 * Try to fill the rest of pagevec using fast
515 * pte walk. This will also update start to
516 * the next page to process. Then munlock the
519 start = __munlock_pagevec_fill(&pvec, vma,
521 __munlock_pagevec(&pvec, zone);
525 /* It's a bug to munlock in the middle of a THP page */
526 VM_BUG_ON((start >> PAGE_SHIFT) & page_mask);
527 page_increm = 1 + page_mask;
528 start += page_increm * PAGE_SIZE;
535 * mlock_fixup - handle mlock[all]/munlock[all] requests.
537 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
538 * munlock is a no-op. However, for some special vmas, we go ahead and
541 * For vmas that pass the filters, merge/split as appropriate.
543 static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
544 unsigned long start, unsigned long end, vm_flags_t newflags)
546 struct mm_struct *mm = vma->vm_mm;
550 int lock = !!(newflags & VM_LOCKED);
552 if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
553 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
554 goto out; /* don't set VM_LOCKED, don't count */
556 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
557 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
558 vma->vm_file, pgoff, vma_policy(vma));
564 if (start != vma->vm_start) {
565 ret = split_vma(mm, vma, start, 1);
570 if (end != vma->vm_end) {
571 ret = split_vma(mm, vma, end, 0);
578 * Keep track of amount of locked VM.
580 nr_pages = (end - start) >> PAGE_SHIFT;
582 nr_pages = -nr_pages;
583 mm->locked_vm += nr_pages;
586 * vm_flags is protected by the mmap_sem held in write mode.
587 * It's okay if try_to_unmap_one unmaps a page just after we
588 * set VM_LOCKED, __mlock_vma_pages_range will bring it back.
592 vma->vm_flags = newflags;
594 munlock_vma_pages_range(vma, start, end);
601 static int do_mlock(unsigned long start, size_t len, int on)
603 unsigned long nstart, end, tmp;
604 struct vm_area_struct * vma, * prev;
607 VM_BUG_ON(start & ~PAGE_MASK);
608 VM_BUG_ON(len != PAGE_ALIGN(len));
614 vma = find_vma(current->mm, start);
615 if (!vma || vma->vm_start > start)
619 if (start > vma->vm_start)
622 for (nstart = start ; ; ) {
625 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
627 newflags = vma->vm_flags & ~VM_LOCKED;
629 newflags |= VM_LOCKED;
634 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
638 if (nstart < prev->vm_end)
639 nstart = prev->vm_end;
644 if (!vma || vma->vm_start != nstart) {
653 * __mm_populate - populate and/or mlock pages within a range of address space.
655 * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap
656 * flags. VMAs must be already marked with the desired vm_flags, and
657 * mmap_sem must not be held.
659 int __mm_populate(unsigned long start, unsigned long len, int ignore_errors)
661 struct mm_struct *mm = current->mm;
662 unsigned long end, nstart, nend;
663 struct vm_area_struct *vma = NULL;
667 VM_BUG_ON(start & ~PAGE_MASK);
668 VM_BUG_ON(len != PAGE_ALIGN(len));
671 for (nstart = start; nstart < end; nstart = nend) {
673 * We want to fault in pages for [nstart; end) address range.
674 * Find first corresponding VMA.
678 down_read(&mm->mmap_sem);
679 vma = find_vma(mm, nstart);
680 } else if (nstart >= vma->vm_end)
682 if (!vma || vma->vm_start >= end)
685 * Set [nstart; nend) to intersection of desired address
686 * range with the first VMA. Also, skip undesirable VMA types.
688 nend = min(end, vma->vm_end);
689 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
691 if (nstart < vma->vm_start)
692 nstart = vma->vm_start;
694 * Now fault in a range of pages. __mlock_vma_pages_range()
695 * double checks the vma flags, so that it won't mlock pages
696 * if the vma was already munlocked.
698 ret = __mlock_vma_pages_range(vma, nstart, nend, &locked);
702 continue; /* continue at next VMA */
704 ret = __mlock_posix_error_return(ret);
707 nend = nstart + ret * PAGE_SIZE;
711 up_read(&mm->mmap_sem);
712 return ret; /* 0 or negative error code */
715 SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
717 unsigned long locked;
718 unsigned long lock_limit;
724 lru_add_drain_all(); /* flush pagevec */
726 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
729 lock_limit = rlimit(RLIMIT_MEMLOCK);
730 lock_limit >>= PAGE_SHIFT;
731 locked = len >> PAGE_SHIFT;
733 down_write(¤t->mm->mmap_sem);
735 locked += current->mm->locked_vm;
737 /* check against resource limits */
738 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
739 error = do_mlock(start, len, 1);
741 up_write(¤t->mm->mmap_sem);
743 error = __mm_populate(start, len, 0);
747 SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
751 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
754 down_write(¤t->mm->mmap_sem);
755 ret = do_mlock(start, len, 0);
756 up_write(¤t->mm->mmap_sem);
761 static int do_mlockall(int flags)
763 struct vm_area_struct * vma, * prev = NULL;
765 if (flags & MCL_FUTURE)
766 current->mm->def_flags |= VM_LOCKED;
768 current->mm->def_flags &= ~VM_LOCKED;
769 if (flags == MCL_FUTURE)
772 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
775 newflags = vma->vm_flags & ~VM_LOCKED;
776 if (flags & MCL_CURRENT)
777 newflags |= VM_LOCKED;
780 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
787 SYSCALL_DEFINE1(mlockall, int, flags)
789 unsigned long lock_limit;
792 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
799 if (flags & MCL_CURRENT)
800 lru_add_drain_all(); /* flush pagevec */
802 lock_limit = rlimit(RLIMIT_MEMLOCK);
803 lock_limit >>= PAGE_SHIFT;
806 down_write(¤t->mm->mmap_sem);
808 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
809 capable(CAP_IPC_LOCK))
810 ret = do_mlockall(flags);
811 up_write(¤t->mm->mmap_sem);
812 if (!ret && (flags & MCL_CURRENT))
813 mm_populate(0, TASK_SIZE);
818 SYSCALL_DEFINE0(munlockall)
822 down_write(¤t->mm->mmap_sem);
823 ret = do_mlockall(0);
824 up_write(¤t->mm->mmap_sem);
829 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
830 * shm segments) get accounted against the user_struct instead.
832 static DEFINE_SPINLOCK(shmlock_user_lock);
834 int user_shm_lock(size_t size, struct user_struct *user)
836 unsigned long lock_limit, locked;
839 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
840 lock_limit = rlimit(RLIMIT_MEMLOCK);
841 if (lock_limit == RLIM_INFINITY)
843 lock_limit >>= PAGE_SHIFT;
844 spin_lock(&shmlock_user_lock);
846 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
849 user->locked_shm += locked;
852 spin_unlock(&shmlock_user_lock);
856 void user_shm_unlock(size_t size, struct user_struct *user)
858 spin_lock(&shmlock_user_lock);
859 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
860 spin_unlock(&shmlock_user_lock);