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 * Finish munlock after successful page isolation
96 * Page must be locked. This is a wrapper for try_to_munlock()
97 * and putback_lru_page() with munlock accounting.
99 static void __munlock_isolated_page(struct page *page)
101 int ret = SWAP_AGAIN;
104 * Optimization: if the page was mapped just once, that's our mapping
105 * and we don't need to check all the other vmas.
107 if (page_mapcount(page) > 1)
108 ret = try_to_munlock(page);
110 /* Did try_to_unlock() succeed or punt? */
111 if (ret != SWAP_MLOCK)
112 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
114 putback_lru_page(page);
118 * Accounting for page isolation fail during munlock
120 * Performs accounting when page isolation fails in munlock. There is nothing
121 * else to do because it means some other task has already removed the page
122 * from the LRU. putback_lru_page() will take care of removing the page from
123 * the unevictable list, if necessary. vmscan [page_referenced()] will move
124 * the page back to the unevictable list if some other vma has it mlocked.
126 static void __munlock_isolation_failed(struct page *page)
128 if (PageUnevictable(page))
129 count_vm_event(UNEVICTABLE_PGSTRANDED);
131 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
135 * munlock_vma_page - munlock a vma page
136 * @page - page to be unlocked
138 * called from munlock()/munmap() path with page supposedly on the LRU.
139 * When we munlock a page, because the vma where we found the page is being
140 * munlock()ed or munmap()ed, we want to check whether other vmas hold the
141 * page locked so that we can leave it on the unevictable lru list and not
142 * bother vmscan with it. However, to walk the page's rmap list in
143 * try_to_munlock() we must isolate the page from the LRU. If some other
144 * task has removed the page from the LRU, we won't be able to do that.
145 * So we clear the PageMlocked as we might not get another chance. If we
146 * can't isolate the page, we leave it for putback_lru_page() and vmscan
147 * [page_referenced()/try_to_unmap()] to deal with.
149 unsigned int munlock_vma_page(struct page *page)
151 unsigned int page_mask = 0;
153 BUG_ON(!PageLocked(page));
155 if (TestClearPageMlocked(page)) {
156 unsigned int nr_pages = hpage_nr_pages(page);
157 mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
158 page_mask = nr_pages - 1;
159 if (!isolate_lru_page(page))
160 __munlock_isolated_page(page);
162 __munlock_isolation_failed(page);
169 * __mlock_vma_pages_range() - mlock a range of pages in the vma.
171 * @start: start address
174 * This takes care of making the pages present too.
176 * return 0 on success, negative error code on error.
178 * vma->vm_mm->mmap_sem must be held for at least read.
180 long __mlock_vma_pages_range(struct vm_area_struct *vma,
181 unsigned long start, unsigned long end, int *nonblocking)
183 struct mm_struct *mm = vma->vm_mm;
184 unsigned long nr_pages = (end - start) / PAGE_SIZE;
187 VM_BUG_ON(start & ~PAGE_MASK);
188 VM_BUG_ON(end & ~PAGE_MASK);
189 VM_BUG_ON(start < vma->vm_start);
190 VM_BUG_ON(end > vma->vm_end);
191 VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem));
193 gup_flags = FOLL_TOUCH | FOLL_MLOCK;
195 * We want to touch writable mappings with a write fault in order
196 * to break COW, except for shared mappings because these don't COW
197 * and we would not want to dirty them for nothing.
199 if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
200 gup_flags |= FOLL_WRITE;
203 * We want mlock to succeed for regions that have any permissions
204 * other than PROT_NONE.
206 if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))
207 gup_flags |= FOLL_FORCE;
210 * We made sure addr is within a VMA, so the following will
211 * not result in a stack expansion that recurses back here.
213 return __get_user_pages(current, mm, start, nr_pages, gup_flags,
214 NULL, NULL, nonblocking);
218 * convert get_user_pages() return value to posix mlock() error
220 static int __mlock_posix_error_return(long retval)
222 if (retval == -EFAULT)
224 else if (retval == -ENOMEM)
230 * Munlock a batch of pages from the same zone
232 * The work is split to two main phases. First phase clears the Mlocked flag
233 * and attempts to isolate the pages, all under a single zone lru lock.
234 * The second phase finishes the munlock only for pages where isolation
237 * Note that pvec is modified during the process. Before returning
238 * pagevec_reinit() is called on it.
240 static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone)
243 int nr = pagevec_count(pvec);
245 /* Phase 1: page isolation */
246 spin_lock_irq(&zone->lru_lock);
247 for (i = 0; i < nr; i++) {
248 struct page *page = pvec->pages[i];
250 if (TestClearPageMlocked(page)) {
251 struct lruvec *lruvec;
254 /* we have disabled interrupts */
255 __mod_zone_page_state(zone, NR_MLOCK, -1);
258 lruvec = mem_cgroup_page_lruvec(page, zone);
259 lru = page_lru(page);
263 del_page_from_lru_list(page, lruvec, lru);
265 __munlock_isolation_failed(page);
272 * We won't be munlocking this page in the next phase
273 * but we still need to release the follow_page_mask()
276 pvec->pages[i] = NULL;
280 spin_unlock_irq(&zone->lru_lock);
282 /* Phase 2: page munlock and putback */
283 for (i = 0; i < nr; i++) {
284 struct page *page = pvec->pages[i];
288 __munlock_isolated_page(page);
290 put_page(page); /* pin from follow_page_mask() */
293 pagevec_reinit(pvec);
297 * munlock_vma_pages_range() - munlock all pages in the vma range.'
298 * @vma - vma containing range to be munlock()ed.
299 * @start - start address in @vma of the range
300 * @end - end of range in @vma.
302 * For mremap(), munmap() and exit().
304 * Called with @vma VM_LOCKED.
306 * Returns with VM_LOCKED cleared. Callers must be prepared to
309 * We don't save and restore VM_LOCKED here because pages are
310 * still on lru. In unmap path, pages might be scanned by reclaim
311 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
312 * free them. This will result in freeing mlocked pages.
314 void munlock_vma_pages_range(struct vm_area_struct *vma,
315 unsigned long start, unsigned long end)
318 struct zone *zone = NULL;
320 pagevec_init(&pvec, 0);
321 vma->vm_flags &= ~VM_LOCKED;
323 while (start < end) {
325 unsigned int page_mask, page_increm;
326 struct zone *pagezone;
329 * Although FOLL_DUMP is intended for get_dump_page(),
330 * it just so happens that its special treatment of the
331 * ZERO_PAGE (returning an error instead of doing get_page)
332 * suits munlock very well (and if somehow an abnormal page
333 * has sneaked into the range, we won't oops here: great).
335 page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP,
337 if (page && !IS_ERR(page)) {
338 pagezone = page_zone(page);
339 /* The whole pagevec must be in the same zone */
340 if (pagezone != zone) {
341 if (pagevec_count(&pvec))
342 __munlock_pagevec(&pvec, zone);
345 if (PageTransHuge(page)) {
347 * THP pages are not handled by pagevec due
348 * to their possible split (see below).
350 if (pagevec_count(&pvec))
351 __munlock_pagevec(&pvec, zone);
354 * Any THP page found by follow_page_mask() may
355 * have gotten split before reaching
356 * munlock_vma_page(), so we need to recompute
357 * the page_mask here.
359 page_mask = munlock_vma_page(page);
361 put_page(page); /* follow_page_mask() */
364 * Non-huge pages are handled in batches
365 * via pagevec. The pin from
366 * follow_page_mask() prevents them from
369 if (pagevec_add(&pvec, page) == 0)
370 __munlock_pagevec(&pvec, zone);
373 page_increm = 1 + (~(start >> PAGE_SHIFT) & page_mask);
374 start += page_increm * PAGE_SIZE;
377 if (pagevec_count(&pvec))
378 __munlock_pagevec(&pvec, zone);
382 * mlock_fixup - handle mlock[all]/munlock[all] requests.
384 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
385 * munlock is a no-op. However, for some special vmas, we go ahead and
388 * For vmas that pass the filters, merge/split as appropriate.
390 static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
391 unsigned long start, unsigned long end, vm_flags_t newflags)
393 struct mm_struct *mm = vma->vm_mm;
397 int lock = !!(newflags & VM_LOCKED);
399 if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
400 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
401 goto out; /* don't set VM_LOCKED, don't count */
403 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
404 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
405 vma->vm_file, pgoff, vma_policy(vma));
411 if (start != vma->vm_start) {
412 ret = split_vma(mm, vma, start, 1);
417 if (end != vma->vm_end) {
418 ret = split_vma(mm, vma, end, 0);
425 * Keep track of amount of locked VM.
427 nr_pages = (end - start) >> PAGE_SHIFT;
429 nr_pages = -nr_pages;
430 mm->locked_vm += nr_pages;
433 * vm_flags is protected by the mmap_sem held in write mode.
434 * It's okay if try_to_unmap_one unmaps a page just after we
435 * set VM_LOCKED, __mlock_vma_pages_range will bring it back.
439 vma->vm_flags = newflags;
441 munlock_vma_pages_range(vma, start, end);
448 static int do_mlock(unsigned long start, size_t len, int on)
450 unsigned long nstart, end, tmp;
451 struct vm_area_struct * vma, * prev;
454 VM_BUG_ON(start & ~PAGE_MASK);
455 VM_BUG_ON(len != PAGE_ALIGN(len));
461 vma = find_vma(current->mm, start);
462 if (!vma || vma->vm_start > start)
466 if (start > vma->vm_start)
469 for (nstart = start ; ; ) {
472 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
474 newflags = vma->vm_flags & ~VM_LOCKED;
476 newflags |= VM_LOCKED;
481 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
485 if (nstart < prev->vm_end)
486 nstart = prev->vm_end;
491 if (!vma || vma->vm_start != nstart) {
500 * __mm_populate - populate and/or mlock pages within a range of address space.
502 * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap
503 * flags. VMAs must be already marked with the desired vm_flags, and
504 * mmap_sem must not be held.
506 int __mm_populate(unsigned long start, unsigned long len, int ignore_errors)
508 struct mm_struct *mm = current->mm;
509 unsigned long end, nstart, nend;
510 struct vm_area_struct *vma = NULL;
514 VM_BUG_ON(start & ~PAGE_MASK);
515 VM_BUG_ON(len != PAGE_ALIGN(len));
518 for (nstart = start; nstart < end; nstart = nend) {
520 * We want to fault in pages for [nstart; end) address range.
521 * Find first corresponding VMA.
525 down_read(&mm->mmap_sem);
526 vma = find_vma(mm, nstart);
527 } else if (nstart >= vma->vm_end)
529 if (!vma || vma->vm_start >= end)
532 * Set [nstart; nend) to intersection of desired address
533 * range with the first VMA. Also, skip undesirable VMA types.
535 nend = min(end, vma->vm_end);
536 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
538 if (nstart < vma->vm_start)
539 nstart = vma->vm_start;
541 * Now fault in a range of pages. __mlock_vma_pages_range()
542 * double checks the vma flags, so that it won't mlock pages
543 * if the vma was already munlocked.
545 ret = __mlock_vma_pages_range(vma, nstart, nend, &locked);
549 continue; /* continue at next VMA */
551 ret = __mlock_posix_error_return(ret);
554 nend = nstart + ret * PAGE_SIZE;
558 up_read(&mm->mmap_sem);
559 return ret; /* 0 or negative error code */
562 SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
564 unsigned long locked;
565 unsigned long lock_limit;
571 lru_add_drain_all(); /* flush pagevec */
573 down_write(¤t->mm->mmap_sem);
574 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
577 locked = len >> PAGE_SHIFT;
578 locked += current->mm->locked_vm;
580 lock_limit = rlimit(RLIMIT_MEMLOCK);
581 lock_limit >>= PAGE_SHIFT;
583 /* check against resource limits */
584 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
585 error = do_mlock(start, len, 1);
586 up_write(¤t->mm->mmap_sem);
588 error = __mm_populate(start, len, 0);
592 SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
596 down_write(¤t->mm->mmap_sem);
597 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
599 ret = do_mlock(start, len, 0);
600 up_write(¤t->mm->mmap_sem);
604 static int do_mlockall(int flags)
606 struct vm_area_struct * vma, * prev = NULL;
608 if (flags & MCL_FUTURE)
609 current->mm->def_flags |= VM_LOCKED;
611 current->mm->def_flags &= ~VM_LOCKED;
612 if (flags == MCL_FUTURE)
615 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
618 newflags = vma->vm_flags & ~VM_LOCKED;
619 if (flags & MCL_CURRENT)
620 newflags |= VM_LOCKED;
623 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
629 SYSCALL_DEFINE1(mlockall, int, flags)
631 unsigned long lock_limit;
634 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
641 if (flags & MCL_CURRENT)
642 lru_add_drain_all(); /* flush pagevec */
644 down_write(¤t->mm->mmap_sem);
646 lock_limit = rlimit(RLIMIT_MEMLOCK);
647 lock_limit >>= PAGE_SHIFT;
650 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
651 capable(CAP_IPC_LOCK))
652 ret = do_mlockall(flags);
653 up_write(¤t->mm->mmap_sem);
654 if (!ret && (flags & MCL_CURRENT))
655 mm_populate(0, TASK_SIZE);
660 SYSCALL_DEFINE0(munlockall)
664 down_write(¤t->mm->mmap_sem);
665 ret = do_mlockall(0);
666 up_write(¤t->mm->mmap_sem);
671 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
672 * shm segments) get accounted against the user_struct instead.
674 static DEFINE_SPINLOCK(shmlock_user_lock);
676 int user_shm_lock(size_t size, struct user_struct *user)
678 unsigned long lock_limit, locked;
681 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
682 lock_limit = rlimit(RLIMIT_MEMLOCK);
683 if (lock_limit == RLIM_INFINITY)
685 lock_limit >>= PAGE_SHIFT;
686 spin_lock(&shmlock_user_lock);
688 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
691 user->locked_shm += locked;
694 spin_unlock(&shmlock_user_lock);
698 void user_shm_unlock(size_t size, struct user_struct *user)
700 spin_lock(&shmlock_user_lock);
701 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
702 spin_unlock(&shmlock_user_lock);