1 // SPDX-License-Identifier: GPL-2.0
5 * (C) Copyright 1995 Linus Torvalds
6 * (C) Copyright 2002 Christoph Hellwig
9 #include <linux/capability.h>
10 #include <linux/mman.h>
12 #include <linux/sched/user.h>
13 #include <linux/swap.h>
14 #include <linux/swapops.h>
15 #include <linux/pagemap.h>
16 #include <linux/pagevec.h>
17 #include <linux/pagewalk.h>
18 #include <linux/mempolicy.h>
19 #include <linux/syscalls.h>
20 #include <linux/sched.h>
21 #include <linux/export.h>
22 #include <linux/rmap.h>
23 #include <linux/mmzone.h>
24 #include <linux/hugetlb.h>
25 #include <linux/memcontrol.h>
26 #include <linux/mm_inline.h>
27 #include <linux/secretmem.h>
31 static DEFINE_PER_CPU(struct pagevec, mlock_pvec);
33 bool can_do_mlock(void)
35 if (rlimit(RLIMIT_MEMLOCK) != 0)
37 if (capable(CAP_IPC_LOCK))
41 EXPORT_SYMBOL(can_do_mlock);
44 * Mlocked pages are marked with PageMlocked() flag for efficient testing
45 * in vmscan and, possibly, the fault path; and to support semi-accurate
48 * An mlocked page [PageMlocked(page)] is unevictable. As such, it will
49 * be placed on the LRU "unevictable" list, rather than the [in]active lists.
50 * The unevictable list is an LRU sibling list to the [in]active lists.
51 * PageUnevictable is set to indicate the unevictable state.
54 static struct lruvec *__mlock_page(struct page *page, struct lruvec *lruvec)
56 /* There is nothing more we can do while it's off LRU */
57 if (!TestClearPageLRU(page))
60 lruvec = folio_lruvec_relock_irq(page_folio(page), lruvec);
62 if (unlikely(page_evictable(page))) {
64 * This is a little surprising, but quite possible:
65 * PageMlocked must have got cleared already by another CPU.
66 * Could this page be on the Unevictable LRU? I'm not sure,
67 * but move it now if so.
69 if (PageUnevictable(page)) {
70 del_page_from_lru_list(page, lruvec);
71 ClearPageUnevictable(page);
72 add_page_to_lru_list(page, lruvec);
73 __count_vm_events(UNEVICTABLE_PGRESCUED,
79 if (PageUnevictable(page)) {
80 if (PageMlocked(page))
85 del_page_from_lru_list(page, lruvec);
86 ClearPageActive(page);
87 SetPageUnevictable(page);
88 page->mlock_count = !!PageMlocked(page);
89 add_page_to_lru_list(page, lruvec);
90 __count_vm_events(UNEVICTABLE_PGCULLED, thp_nr_pages(page));
96 static struct lruvec *__mlock_new_page(struct page *page, struct lruvec *lruvec)
98 VM_BUG_ON_PAGE(PageLRU(page), page);
100 lruvec = folio_lruvec_relock_irq(page_folio(page), lruvec);
102 /* As above, this is a little surprising, but possible */
103 if (unlikely(page_evictable(page)))
106 SetPageUnevictable(page);
107 page->mlock_count = !!PageMlocked(page);
108 __count_vm_events(UNEVICTABLE_PGCULLED, thp_nr_pages(page));
110 add_page_to_lru_list(page, lruvec);
115 static struct lruvec *__munlock_page(struct page *page, struct lruvec *lruvec)
117 int nr_pages = thp_nr_pages(page);
118 bool isolated = false;
120 if (!TestClearPageLRU(page))
124 lruvec = folio_lruvec_relock_irq(page_folio(page), lruvec);
126 if (PageUnevictable(page)) {
127 /* Then mlock_count is maintained, but might undercount */
128 if (page->mlock_count)
130 if (page->mlock_count)
133 /* else assume that was the last mlock: reclaim will fix it if not */
136 if (TestClearPageMlocked(page)) {
137 __mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
138 if (isolated || !PageUnevictable(page))
139 __count_vm_events(UNEVICTABLE_PGMUNLOCKED, nr_pages);
141 __count_vm_events(UNEVICTABLE_PGSTRANDED, nr_pages);
144 /* page_evictable() has to be checked *after* clearing Mlocked */
145 if (isolated && PageUnevictable(page) && page_evictable(page)) {
146 del_page_from_lru_list(page, lruvec);
147 ClearPageUnevictable(page);
148 add_page_to_lru_list(page, lruvec);
149 __count_vm_events(UNEVICTABLE_PGRESCUED, nr_pages);
158 * Flags held in the low bits of a struct page pointer on the mlock_pvec.
162 static inline struct page *mlock_lru(struct page *page)
164 return (struct page *)((unsigned long)page + LRU_PAGE);
167 static inline struct page *mlock_new(struct page *page)
169 return (struct page *)((unsigned long)page + NEW_PAGE);
173 * mlock_pagevec() is derived from pagevec_lru_move_fn():
174 * perhaps that can make use of such page pointer flags in future,
175 * but for now just keep it for mlock. We could use three separate
176 * pagevecs instead, but one feels better (munlocking a full pagevec
177 * does not need to drain mlocking pagevecs first).
179 static void mlock_pagevec(struct pagevec *pvec)
181 struct lruvec *lruvec = NULL;
186 for (i = 0; i < pagevec_count(pvec); i++) {
187 page = pvec->pages[i];
188 mlock = (unsigned long)page & (LRU_PAGE | NEW_PAGE);
189 page = (struct page *)((unsigned long)page - mlock);
190 pvec->pages[i] = page;
192 if (mlock & LRU_PAGE)
193 lruvec = __mlock_page(page, lruvec);
194 else if (mlock & NEW_PAGE)
195 lruvec = __mlock_new_page(page, lruvec);
197 lruvec = __munlock_page(page, lruvec);
201 unlock_page_lruvec_irq(lruvec);
202 release_pages(pvec->pages, pvec->nr);
203 pagevec_reinit(pvec);
206 void mlock_page_drain(int cpu)
208 struct pagevec *pvec;
210 pvec = &per_cpu(mlock_pvec, cpu);
211 if (pagevec_count(pvec))
215 bool need_mlock_page_drain(int cpu)
217 return pagevec_count(&per_cpu(mlock_pvec, cpu));
221 * mlock_page - mlock a page already on (or temporarily off) LRU
222 * @page: page to be mlocked, either a normal page or a THP head.
224 void mlock_page(struct page *page)
226 struct pagevec *pvec = &get_cpu_var(mlock_pvec);
228 if (!TestSetPageMlocked(page)) {
229 int nr_pages = thp_nr_pages(page);
231 mod_zone_page_state(page_zone(page), NR_MLOCK, nr_pages);
232 __count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
236 if (!pagevec_add(pvec, mlock_lru(page)) ||
237 PageHead(page) || lru_cache_disabled())
239 put_cpu_var(mlock_pvec);
243 * mlock_new_page - mlock a newly allocated page not yet on LRU
244 * @page: page to be mlocked, either a normal page or a THP head.
246 void mlock_new_page(struct page *page)
248 struct pagevec *pvec = &get_cpu_var(mlock_pvec);
249 int nr_pages = thp_nr_pages(page);
251 SetPageMlocked(page);
252 mod_zone_page_state(page_zone(page), NR_MLOCK, nr_pages);
253 __count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
256 if (!pagevec_add(pvec, mlock_new(page)) ||
257 PageHead(page) || lru_cache_disabled())
259 put_cpu_var(mlock_pvec);
263 * munlock_page - munlock a page
264 * @page: page to be munlocked, either a normal page or a THP head.
266 void munlock_page(struct page *page)
268 struct pagevec *pvec = &get_cpu_var(mlock_pvec);
271 * TestClearPageMlocked(page) must be left to __munlock_page(),
272 * which will check whether the page is multiply mlocked.
276 if (!pagevec_add(pvec, page) ||
277 PageHead(page) || lru_cache_disabled())
279 put_cpu_var(mlock_pvec);
282 static int mlock_pte_range(pmd_t *pmd, unsigned long addr,
283 unsigned long end, struct mm_walk *walk)
286 struct vm_area_struct *vma = walk->vma;
288 pte_t *start_pte, *pte;
291 ptl = pmd_trans_huge_lock(pmd, vma);
293 if (!pmd_present(*pmd))
295 if (is_huge_zero_pmd(*pmd))
297 page = pmd_page(*pmd);
298 if (vma->vm_flags & VM_LOCKED)
305 start_pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
306 for (pte = start_pte; addr != end; pte++, addr += PAGE_SIZE) {
307 if (!pte_present(*pte))
309 page = vm_normal_page(vma, addr, *pte);
312 if (PageTransCompound(page))
314 if (vma->vm_flags & VM_LOCKED)
319 pte_unmap(start_pte);
327 * mlock_vma_pages_range() - mlock any pages already in the range,
328 * or munlock all pages in the range.
329 * @vma - vma containing range to be mlock()ed or munlock()ed
330 * @start - start address in @vma of the range
331 * @end - end of range in @vma
332 * @newflags - the new set of flags for @vma.
334 * Called for mlock(), mlock2() and mlockall(), to set @vma VM_LOCKED;
335 * called for munlock() and munlockall(), to clear VM_LOCKED from @vma.
337 static void mlock_vma_pages_range(struct vm_area_struct *vma,
338 unsigned long start, unsigned long end, vm_flags_t newflags)
340 static const struct mm_walk_ops mlock_walk_ops = {
341 .pmd_entry = mlock_pte_range,
345 * There is a slight chance that concurrent page migration,
346 * or page reclaim finding a page of this now-VM_LOCKED vma,
347 * will call mlock_vma_page() and raise page's mlock_count:
348 * double counting, leaving the page unevictable indefinitely.
349 * Communicate this danger to mlock_vma_page() with VM_IO,
350 * which is a VM_SPECIAL flag not allowed on VM_LOCKED vmas.
351 * mmap_lock is held in write mode here, so this weird
352 * combination should not be visible to other mmap_lock users;
353 * but WRITE_ONCE so rmap walkers must see VM_IO if VM_LOCKED.
355 if (newflags & VM_LOCKED)
357 WRITE_ONCE(vma->vm_flags, newflags);
360 walk_page_range(vma->vm_mm, start, end, &mlock_walk_ops, NULL);
363 if (newflags & VM_IO) {
365 WRITE_ONCE(vma->vm_flags, newflags);
370 * mlock_fixup - handle mlock[all]/munlock[all] requests.
372 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
373 * munlock is a no-op. However, for some special vmas, we go ahead and
376 * For vmas that pass the filters, merge/split as appropriate.
378 static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
379 unsigned long start, unsigned long end, vm_flags_t newflags)
381 struct mm_struct *mm = vma->vm_mm;
385 vm_flags_t oldflags = vma->vm_flags;
387 if (newflags == oldflags || (oldflags & VM_SPECIAL) ||
388 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm) ||
389 vma_is_dax(vma) || vma_is_secretmem(vma))
390 /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
393 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
394 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
395 vma->vm_file, pgoff, vma_policy(vma),
396 vma->vm_userfaultfd_ctx, vma_anon_name(vma));
402 if (start != vma->vm_start) {
403 ret = split_vma(mm, vma, start, 1);
408 if (end != vma->vm_end) {
409 ret = split_vma(mm, vma, end, 0);
416 * Keep track of amount of locked VM.
418 nr_pages = (end - start) >> PAGE_SHIFT;
419 if (!(newflags & VM_LOCKED))
420 nr_pages = -nr_pages;
421 else if (oldflags & VM_LOCKED)
423 mm->locked_vm += nr_pages;
426 * vm_flags is protected by the mmap_lock held in write mode.
427 * It's okay if try_to_unmap_one unmaps a page just after we
428 * set VM_LOCKED, populate_vma_page_range will bring it back.
431 if ((newflags & VM_LOCKED) && (oldflags & VM_LOCKED)) {
432 /* No work to do, and mlocking twice would be wrong */
433 vma->vm_flags = newflags;
435 mlock_vma_pages_range(vma, start, end, newflags);
442 static int apply_vma_lock_flags(unsigned long start, size_t len,
445 unsigned long nstart, end, tmp;
446 struct vm_area_struct *vma, *prev;
449 VM_BUG_ON(offset_in_page(start));
450 VM_BUG_ON(len != PAGE_ALIGN(len));
456 vma = find_vma(current->mm, start);
457 if (!vma || vma->vm_start > start)
461 if (start > vma->vm_start)
464 for (nstart = start ; ; ) {
465 vm_flags_t newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
469 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
473 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
477 if (nstart < prev->vm_end)
478 nstart = prev->vm_end;
483 if (!vma || vma->vm_start != nstart) {
492 * Go through vma areas and sum size of mlocked
493 * vma pages, as return value.
494 * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT)
496 * Return value: previously mlocked page counts
498 static unsigned long count_mm_mlocked_page_nr(struct mm_struct *mm,
499 unsigned long start, size_t len)
501 struct vm_area_struct *vma;
502 unsigned long count = 0;
507 vma = find_vma(mm, start);
511 for (; vma ; vma = vma->vm_next) {
512 if (start >= vma->vm_end)
514 if (start + len <= vma->vm_start)
516 if (vma->vm_flags & VM_LOCKED) {
517 if (start > vma->vm_start)
518 count -= (start - vma->vm_start);
519 if (start + len < vma->vm_end) {
520 count += start + len - vma->vm_start;
523 count += vma->vm_end - vma->vm_start;
527 return count >> PAGE_SHIFT;
531 * convert get_user_pages() return value to posix mlock() error
533 static int __mlock_posix_error_return(long retval)
535 if (retval == -EFAULT)
537 else if (retval == -ENOMEM)
542 static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
544 unsigned long locked;
545 unsigned long lock_limit;
548 start = untagged_addr(start);
553 len = PAGE_ALIGN(len + (offset_in_page(start)));
556 lock_limit = rlimit(RLIMIT_MEMLOCK);
557 lock_limit >>= PAGE_SHIFT;
558 locked = len >> PAGE_SHIFT;
560 if (mmap_write_lock_killable(current->mm))
563 locked += current->mm->locked_vm;
564 if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) {
566 * It is possible that the regions requested intersect with
567 * previously mlocked areas, that part area in "mm->locked_vm"
568 * should not be counted to new mlock increment count. So check
569 * and adjust locked count if necessary.
571 locked -= count_mm_mlocked_page_nr(current->mm,
575 /* check against resource limits */
576 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
577 error = apply_vma_lock_flags(start, len, flags);
579 mmap_write_unlock(current->mm);
583 error = __mm_populate(start, len, 0);
585 return __mlock_posix_error_return(error);
589 SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
591 return do_mlock(start, len, VM_LOCKED);
594 SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
596 vm_flags_t vm_flags = VM_LOCKED;
598 if (flags & ~MLOCK_ONFAULT)
601 if (flags & MLOCK_ONFAULT)
602 vm_flags |= VM_LOCKONFAULT;
604 return do_mlock(start, len, vm_flags);
607 SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
611 start = untagged_addr(start);
613 len = PAGE_ALIGN(len + (offset_in_page(start)));
616 if (mmap_write_lock_killable(current->mm))
618 ret = apply_vma_lock_flags(start, len, 0);
619 mmap_write_unlock(current->mm);
625 * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
626 * and translate into the appropriate modifications to mm->def_flags and/or the
627 * flags for all current VMAs.
629 * There are a couple of subtleties with this. If mlockall() is called multiple
630 * times with different flags, the values do not necessarily stack. If mlockall
631 * is called once including the MCL_FUTURE flag and then a second time without
632 * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
634 static int apply_mlockall_flags(int flags)
636 struct vm_area_struct *vma, *prev = NULL;
637 vm_flags_t to_add = 0;
639 current->mm->def_flags &= VM_LOCKED_CLEAR_MASK;
640 if (flags & MCL_FUTURE) {
641 current->mm->def_flags |= VM_LOCKED;
643 if (flags & MCL_ONFAULT)
644 current->mm->def_flags |= VM_LOCKONFAULT;
646 if (!(flags & MCL_CURRENT))
650 if (flags & MCL_CURRENT) {
652 if (flags & MCL_ONFAULT)
653 to_add |= VM_LOCKONFAULT;
656 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
659 newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
663 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
670 SYSCALL_DEFINE1(mlockall, int, flags)
672 unsigned long lock_limit;
675 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)) ||
676 flags == MCL_ONFAULT)
682 lock_limit = rlimit(RLIMIT_MEMLOCK);
683 lock_limit >>= PAGE_SHIFT;
685 if (mmap_write_lock_killable(current->mm))
689 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
690 capable(CAP_IPC_LOCK))
691 ret = apply_mlockall_flags(flags);
692 mmap_write_unlock(current->mm);
693 if (!ret && (flags & MCL_CURRENT))
694 mm_populate(0, TASK_SIZE);
699 SYSCALL_DEFINE0(munlockall)
703 if (mmap_write_lock_killable(current->mm))
705 ret = apply_mlockall_flags(0);
706 mmap_write_unlock(current->mm);
711 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
712 * shm segments) get accounted against the user_struct instead.
714 static DEFINE_SPINLOCK(shmlock_user_lock);
716 int user_shm_lock(size_t size, struct ucounts *ucounts)
718 unsigned long lock_limit, locked;
722 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
723 lock_limit = rlimit(RLIMIT_MEMLOCK);
724 if (lock_limit == RLIM_INFINITY)
726 lock_limit >>= PAGE_SHIFT;
727 spin_lock(&shmlock_user_lock);
728 memlock = inc_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
730 if (!allowed && (memlock == LONG_MAX || memlock > lock_limit) && !capable(CAP_IPC_LOCK)) {
731 dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
734 if (!get_ucounts(ucounts)) {
735 dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
740 spin_unlock(&shmlock_user_lock);
744 void user_shm_unlock(size_t size, struct ucounts *ucounts)
746 spin_lock(&shmlock_user_lock);
747 dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, (size + PAGE_SIZE - 1) >> PAGE_SHIFT);
748 spin_unlock(&shmlock_user_lock);
749 put_ucounts(ucounts);