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/mempolicy.h>
15 #include <linux/syscalls.h>
16 #include <linux/sched.h>
17 #include <linux/export.h>
18 #include <linux/rmap.h>
19 #include <linux/mmzone.h>
20 #include <linux/hugetlb.h>
24 int can_do_mlock(void)
26 if (capable(CAP_IPC_LOCK))
28 if (rlimit(RLIMIT_MEMLOCK) != 0)
32 EXPORT_SYMBOL(can_do_mlock);
35 * Mlocked pages are marked with PageMlocked() flag for efficient testing
36 * in vmscan and, possibly, the fault path; and to support semi-accurate
39 * An mlocked page [PageMlocked(page)] is unevictable. As such, it will
40 * be placed on the LRU "unevictable" list, rather than the [in]active lists.
41 * The unevictable list is an LRU sibling list to the [in]active lists.
42 * PageUnevictable is set to indicate the unevictable state.
44 * When lazy mlocking via vmscan, it is important to ensure that the
45 * vma's VM_LOCKED status is not concurrently being modified, otherwise we
46 * may have mlocked a page that is being munlocked. So lazy mlock must take
47 * the mmap_sem for read, and verify that the vma really is locked
52 * LRU accounting for clear_page_mlock()
54 void clear_page_mlock(struct page *page)
56 if (!TestClearPageMlocked(page))
59 mod_zone_page_state(page_zone(page), NR_MLOCK,
60 -hpage_nr_pages(page));
61 count_vm_event(UNEVICTABLE_PGCLEARED);
62 if (!isolate_lru_page(page)) {
63 putback_lru_page(page);
66 * We lost the race. the page already moved to evictable list.
68 if (PageUnevictable(page))
69 count_vm_event(UNEVICTABLE_PGSTRANDED);
74 * Mark page as mlocked if not already.
75 * If page on LRU, isolate and putback to move to unevictable list.
77 void mlock_vma_page(struct page *page)
79 /* Serialize with page migration */
80 BUG_ON(!PageLocked(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 * munlock_vma_page - munlock a vma page
93 * @page - page to be unlocked
95 * called from munlock()/munmap() path with page supposedly on the LRU.
96 * When we munlock a page, because the vma where we found the page is being
97 * munlock()ed or munmap()ed, we want to check whether other vmas hold the
98 * page locked so that we can leave it on the unevictable lru list and not
99 * bother vmscan with it. However, to walk the page's rmap list in
100 * try_to_munlock() we must isolate the page from the LRU. If some other
101 * task has removed the page from the LRU, we won't be able to do that.
102 * So we clear the PageMlocked as we might not get another chance. If we
103 * can't isolate the page, we leave it for putback_lru_page() and vmscan
104 * [page_referenced()/try_to_unmap()] to deal with.
106 unsigned int munlock_vma_page(struct page *page)
108 unsigned int page_mask = 0;
110 /* For try_to_munlock() and to serialize with page migration */
111 BUG_ON(!PageLocked(page));
113 if (TestClearPageMlocked(page)) {
114 unsigned int nr_pages = hpage_nr_pages(page);
115 mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
116 page_mask = nr_pages - 1;
117 if (!isolate_lru_page(page)) {
118 int ret = SWAP_AGAIN;
121 * Optimization: if the page was mapped just once,
122 * that's our mapping and we don't need to check all the
125 if (page_mapcount(page) > 1)
126 ret = try_to_munlock(page);
128 * did try_to_unlock() succeed or punt?
130 if (ret != SWAP_MLOCK)
131 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
133 putback_lru_page(page);
136 * Some other task has removed the page from the LRU.
137 * putback_lru_page() will take care of removing the
138 * page from the unevictable list, if necessary.
139 * vmscan [page_referenced()] will move the page back
140 * to the unevictable list if some other vma has it
143 if (PageUnevictable(page))
144 count_vm_event(UNEVICTABLE_PGSTRANDED);
146 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
154 * __mlock_vma_pages_range() - mlock a range of pages in the vma.
156 * @start: start address
159 * This takes care of making the pages present too.
161 * return 0 on success, negative error code on error.
163 * vma->vm_mm->mmap_sem must be held for at least read.
165 long __mlock_vma_pages_range(struct vm_area_struct *vma,
166 unsigned long start, unsigned long end, int *nonblocking)
168 struct mm_struct *mm = vma->vm_mm;
169 unsigned long nr_pages = (end - start) / PAGE_SIZE;
172 VM_BUG_ON(start & ~PAGE_MASK);
173 VM_BUG_ON(end & ~PAGE_MASK);
174 VM_BUG_ON(start < vma->vm_start);
175 VM_BUG_ON(end > vma->vm_end);
176 VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem));
178 gup_flags = FOLL_TOUCH | FOLL_MLOCK;
180 * We want to touch writable mappings with a write fault in order
181 * to break COW, except for shared mappings because these don't COW
182 * and we would not want to dirty them for nothing.
184 if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
185 gup_flags |= FOLL_WRITE;
188 * We want mlock to succeed for regions that have any permissions
189 * other than PROT_NONE.
191 if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))
192 gup_flags |= FOLL_FORCE;
195 * We made sure addr is within a VMA, so the following will
196 * not result in a stack expansion that recurses back here.
198 return __get_user_pages(current, mm, start, nr_pages, gup_flags,
199 NULL, NULL, nonblocking);
203 * convert get_user_pages() return value to posix mlock() error
205 static int __mlock_posix_error_return(long retval)
207 if (retval == -EFAULT)
209 else if (retval == -ENOMEM)
215 * munlock_vma_pages_range() - munlock all pages in the vma range.'
216 * @vma - vma containing range to be munlock()ed.
217 * @start - start address in @vma of the range
218 * @end - end of range in @vma.
220 * For mremap(), munmap() and exit().
222 * Called with @vma VM_LOCKED.
224 * Returns with VM_LOCKED cleared. Callers must be prepared to
227 * We don't save and restore VM_LOCKED here because pages are
228 * still on lru. In unmap path, pages might be scanned by reclaim
229 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
230 * free them. This will result in freeing mlocked pages.
232 void munlock_vma_pages_range(struct vm_area_struct *vma,
233 unsigned long start, unsigned long end)
235 vma->vm_flags &= ~VM_LOCKED;
237 while (start < end) {
239 unsigned int page_mask, page_increm;
242 * Although FOLL_DUMP is intended for get_dump_page(),
243 * it just so happens that its special treatment of the
244 * ZERO_PAGE (returning an error instead of doing get_page)
245 * suits munlock very well (and if somehow an abnormal page
246 * has sneaked into the range, we won't oops here: great).
248 page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP,
250 if (page && !IS_ERR(page)) {
254 * Any THP page found by follow_page_mask() may have
255 * gotten split before reaching munlock_vma_page(),
256 * so we need to recompute the page_mask here.
258 page_mask = munlock_vma_page(page);
262 page_increm = 1 + (~(start >> PAGE_SHIFT) & page_mask);
263 start += page_increm * PAGE_SIZE;
269 * mlock_fixup - handle mlock[all]/munlock[all] requests.
271 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
272 * munlock is a no-op. However, for some special vmas, we go ahead and
275 * For vmas that pass the filters, merge/split as appropriate.
277 static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
278 unsigned long start, unsigned long end, vm_flags_t newflags)
280 struct mm_struct *mm = vma->vm_mm;
284 int lock = !!(newflags & VM_LOCKED);
286 if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
287 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
288 goto out; /* don't set VM_LOCKED, don't count */
290 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
291 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
292 vma->vm_file, pgoff, vma_policy(vma),
293 vma_get_anon_name(vma));
299 if (start != vma->vm_start) {
300 ret = split_vma(mm, vma, start, 1);
305 if (end != vma->vm_end) {
306 ret = split_vma(mm, vma, end, 0);
313 * Keep track of amount of locked VM.
315 nr_pages = (end - start) >> PAGE_SHIFT;
317 nr_pages = -nr_pages;
318 mm->locked_vm += nr_pages;
321 * vm_flags is protected by the mmap_sem held in write mode.
322 * It's okay if try_to_unmap_one unmaps a page just after we
323 * set VM_LOCKED, __mlock_vma_pages_range will bring it back.
327 vma->vm_flags = newflags;
329 munlock_vma_pages_range(vma, start, end);
336 static int do_mlock(unsigned long start, size_t len, int on)
338 unsigned long nstart, end, tmp;
339 struct vm_area_struct * vma, * prev;
342 VM_BUG_ON(start & ~PAGE_MASK);
343 VM_BUG_ON(len != PAGE_ALIGN(len));
349 vma = find_vma(current->mm, start);
350 if (!vma || vma->vm_start > start)
354 if (start > vma->vm_start)
357 for (nstart = start ; ; ) {
360 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
362 newflags = vma->vm_flags & ~VM_LOCKED;
364 newflags |= VM_LOCKED;
369 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
373 if (nstart < prev->vm_end)
374 nstart = prev->vm_end;
379 if (!vma || vma->vm_start != nstart) {
388 * __mm_populate - populate and/or mlock pages within a range of address space.
390 * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap
391 * flags. VMAs must be already marked with the desired vm_flags, and
392 * mmap_sem must not be held.
394 int __mm_populate(unsigned long start, unsigned long len, int ignore_errors)
396 struct mm_struct *mm = current->mm;
397 unsigned long end, nstart, nend;
398 struct vm_area_struct *vma = NULL;
402 VM_BUG_ON(start & ~PAGE_MASK);
403 VM_BUG_ON(len != PAGE_ALIGN(len));
406 for (nstart = start; nstart < end; nstart = nend) {
408 * We want to fault in pages for [nstart; end) address range.
409 * Find first corresponding VMA.
413 down_read(&mm->mmap_sem);
414 vma = find_vma(mm, nstart);
415 } else if (nstart >= vma->vm_end)
417 if (!vma || vma->vm_start >= end)
420 * Set [nstart; nend) to intersection of desired address
421 * range with the first VMA. Also, skip undesirable VMA types.
423 nend = min(end, vma->vm_end);
424 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
426 if (nstart < vma->vm_start)
427 nstart = vma->vm_start;
429 * Now fault in a range of pages. __mlock_vma_pages_range()
430 * double checks the vma flags, so that it won't mlock pages
431 * if the vma was already munlocked.
433 ret = __mlock_vma_pages_range(vma, nstart, nend, &locked);
437 continue; /* continue at next VMA */
439 ret = __mlock_posix_error_return(ret);
442 nend = nstart + ret * PAGE_SIZE;
446 up_read(&mm->mmap_sem);
447 return ret; /* 0 or negative error code */
450 SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
452 unsigned long locked;
453 unsigned long lock_limit;
459 lru_add_drain_all(); /* flush pagevec */
461 down_write(¤t->mm->mmap_sem);
462 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
465 locked = len >> PAGE_SHIFT;
466 locked += current->mm->locked_vm;
468 lock_limit = rlimit(RLIMIT_MEMLOCK);
469 lock_limit >>= PAGE_SHIFT;
471 /* check against resource limits */
472 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
473 error = do_mlock(start, len, 1);
474 up_write(¤t->mm->mmap_sem);
476 error = __mm_populate(start, len, 0);
480 SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
484 down_write(¤t->mm->mmap_sem);
485 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
487 ret = do_mlock(start, len, 0);
488 up_write(¤t->mm->mmap_sem);
492 static int do_mlockall(int flags)
494 struct vm_area_struct * vma, * prev = NULL;
496 if (flags & MCL_FUTURE)
497 current->mm->def_flags |= VM_LOCKED;
499 current->mm->def_flags &= ~VM_LOCKED;
500 if (flags == MCL_FUTURE)
503 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
506 newflags = vma->vm_flags & ~VM_LOCKED;
507 if (flags & MCL_CURRENT)
508 newflags |= VM_LOCKED;
511 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
517 SYSCALL_DEFINE1(mlockall, int, flags)
519 unsigned long lock_limit;
522 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
529 if (flags & MCL_CURRENT)
530 lru_add_drain_all(); /* flush pagevec */
532 down_write(¤t->mm->mmap_sem);
534 lock_limit = rlimit(RLIMIT_MEMLOCK);
535 lock_limit >>= PAGE_SHIFT;
538 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
539 capable(CAP_IPC_LOCK))
540 ret = do_mlockall(flags);
541 up_write(¤t->mm->mmap_sem);
542 if (!ret && (flags & MCL_CURRENT))
543 mm_populate(0, TASK_SIZE);
548 SYSCALL_DEFINE0(munlockall)
552 down_write(¤t->mm->mmap_sem);
553 ret = do_mlockall(0);
554 up_write(¤t->mm->mmap_sem);
559 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
560 * shm segments) get accounted against the user_struct instead.
562 static DEFINE_SPINLOCK(shmlock_user_lock);
564 int user_shm_lock(size_t size, struct user_struct *user)
566 unsigned long lock_limit, locked;
569 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
570 lock_limit = rlimit(RLIMIT_MEMLOCK);
571 if (lock_limit == RLIM_INFINITY)
573 lock_limit >>= PAGE_SHIFT;
574 spin_lock(&shmlock_user_lock);
576 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
579 user->locked_shm += locked;
582 spin_unlock(&shmlock_user_lock);
586 void user_shm_unlock(size_t size, struct user_struct *user)
588 spin_lock(&shmlock_user_lock);
589 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
590 spin_unlock(&shmlock_user_lock);