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/module.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 VM_BUG_ON(!PageLocked(page));
58 if (!page->mapping) { /* truncated ? */
62 dec_zone_page_state(page, NR_MLOCK);
63 count_vm_event(UNEVICTABLE_PGCLEARED);
64 if (!isolate_lru_page(page)) {
65 putback_lru_page(page);
68 * We lost the race. the page already moved to evictable list.
70 if (PageUnevictable(page))
71 count_vm_event(UNEVICTABLE_PGSTRANDED);
76 * Mark page as mlocked if not already.
77 * If page on LRU, isolate and putback to move to unevictable list.
79 void mlock_vma_page(struct page *page)
81 BUG_ON(!PageLocked(page));
83 if (!TestSetPageMlocked(page)) {
84 inc_zone_page_state(page, NR_MLOCK);
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 void munlock_vma_page(struct page *page)
108 BUG_ON(!PageLocked(page));
110 if (TestClearPageMlocked(page)) {
111 dec_zone_page_state(page, NR_MLOCK);
112 if (!isolate_lru_page(page)) {
113 int ret = try_to_munlock(page);
115 * did try_to_unlock() succeed or punt?
117 if (ret != SWAP_MLOCK)
118 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
120 putback_lru_page(page);
123 * Some other task has removed the page from the LRU.
124 * putback_lru_page() will take care of removing the
125 * page from the unevictable list, if necessary.
126 * vmscan [page_referenced()] will move the page back
127 * to the unevictable list if some other vma has it
130 if (PageUnevictable(page))
131 count_vm_event(UNEVICTABLE_PGSTRANDED);
133 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
138 static inline int stack_guard_page(struct vm_area_struct *vma, unsigned long addr)
140 return (vma->vm_flags & VM_GROWSDOWN) &&
141 (vma->vm_start == addr) &&
142 !vma_stack_continue(vma->vm_prev, addr);
146 * __mlock_vma_pages_range() - mlock a range of pages in the vma.
148 * @start: start address
151 * This takes care of making the pages present too.
153 * return 0 on success, negative error code on error.
155 * vma->vm_mm->mmap_sem must be held for at least read.
157 static long __mlock_vma_pages_range(struct vm_area_struct *vma,
158 unsigned long start, unsigned long end,
161 struct mm_struct *mm = vma->vm_mm;
162 unsigned long addr = start;
163 int nr_pages = (end - start) / PAGE_SIZE;
166 VM_BUG_ON(start & ~PAGE_MASK);
167 VM_BUG_ON(end & ~PAGE_MASK);
168 VM_BUG_ON(start < vma->vm_start);
169 VM_BUG_ON(end > vma->vm_end);
170 VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem));
172 gup_flags = FOLL_TOUCH;
174 * We want to touch writable mappings with a write fault in order
175 * to break COW, except for shared mappings because these don't COW
176 * and we would not want to dirty them for nothing.
178 if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
179 gup_flags |= FOLL_WRITE;
182 * We want mlock to succeed for regions that have any permissions
183 * other than PROT_NONE.
185 if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))
186 gup_flags |= FOLL_FORCE;
188 if (vma->vm_flags & VM_LOCKED)
189 gup_flags |= FOLL_MLOCK;
191 /* We don't try to access the guard page of a stack vma */
192 if (stack_guard_page(vma, start)) {
197 return __get_user_pages(current, mm, addr, nr_pages, gup_flags,
198 NULL, NULL, nonblocking);
202 * convert get_user_pages() return value to posix mlock() error
204 static int __mlock_posix_error_return(long retval)
206 if (retval == -EFAULT)
208 else if (retval == -ENOMEM)
214 * mlock_vma_pages_range() - mlock pages in specified vma range.
215 * @vma - the vma containing the specfied address range
216 * @start - starting address in @vma to mlock
217 * @end - end address [+1] in @vma to mlock
219 * For mmap()/mremap()/expansion of mlocked vma.
221 * return 0 on success for "normal" vmas.
223 * return number of pages [> 0] to be removed from locked_vm on success
226 long mlock_vma_pages_range(struct vm_area_struct *vma,
227 unsigned long start, unsigned long end)
229 int nr_pages = (end - start) / PAGE_SIZE;
230 BUG_ON(!(vma->vm_flags & VM_LOCKED));
233 * filter unlockable vmas
235 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
238 if (!((vma->vm_flags & (VM_DONTEXPAND | VM_RESERVED)) ||
239 is_vm_hugetlb_page(vma) ||
240 vma == get_gate_vma(current->mm))) {
242 __mlock_vma_pages_range(vma, start, end, NULL);
244 /* Hide errors from mmap() and other callers */
249 * User mapped kernel pages or huge pages:
250 * make these pages present to populate the ptes, but
251 * fall thru' to reset VM_LOCKED--no need to unlock, and
252 * return nr_pages so these don't get counted against task's
253 * locked limit. huge pages are already counted against
256 make_pages_present(start, end);
259 vma->vm_flags &= ~VM_LOCKED; /* and don't come back! */
260 return nr_pages; /* error or pages NOT mlocked */
264 * munlock_vma_pages_range() - munlock all pages in the vma range.'
265 * @vma - vma containing range to be munlock()ed.
266 * @start - start address in @vma of the range
267 * @end - end of range in @vma.
269 * For mremap(), munmap() and exit().
271 * Called with @vma VM_LOCKED.
273 * Returns with VM_LOCKED cleared. Callers must be prepared to
276 * We don't save and restore VM_LOCKED here because pages are
277 * still on lru. In unmap path, pages might be scanned by reclaim
278 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
279 * free them. This will result in freeing mlocked pages.
281 void munlock_vma_pages_range(struct vm_area_struct *vma,
282 unsigned long start, unsigned long end)
287 vma->vm_flags &= ~VM_LOCKED;
289 for (addr = start; addr < end; addr += PAGE_SIZE) {
292 * Although FOLL_DUMP is intended for get_dump_page(),
293 * it just so happens that its special treatment of the
294 * ZERO_PAGE (returning an error instead of doing get_page)
295 * suits munlock very well (and if somehow an abnormal page
296 * has sneaked into the range, we won't oops here: great).
298 page = follow_page(vma, addr, FOLL_GET | FOLL_DUMP);
299 if (page && !IS_ERR(page)) {
302 * Like in __mlock_vma_pages_range(),
303 * because we lock page here and migration is
304 * blocked by the elevated reference, we need
305 * only check for file-cache page truncation.
308 munlock_vma_page(page);
317 * mlock_fixup - handle mlock[all]/munlock[all] requests.
319 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
320 * munlock is a no-op. However, for some special vmas, we go ahead and
321 * populate the ptes via make_pages_present().
323 * For vmas that pass the filters, merge/split as appropriate.
325 static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
326 unsigned long start, unsigned long end, unsigned int newflags)
328 struct mm_struct *mm = vma->vm_mm;
332 int lock = newflags & VM_LOCKED;
334 if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
335 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
336 goto out; /* don't set VM_LOCKED, don't count */
338 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
339 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
340 vma->vm_file, pgoff, vma_policy(vma));
346 if (start != vma->vm_start) {
347 ret = split_vma(mm, vma, start, 1);
352 if (end != vma->vm_end) {
353 ret = split_vma(mm, vma, end, 0);
360 * Keep track of amount of locked VM.
362 nr_pages = (end - start) >> PAGE_SHIFT;
364 nr_pages = -nr_pages;
365 mm->locked_vm += nr_pages;
368 * vm_flags is protected by the mmap_sem held in write mode.
369 * It's okay if try_to_unmap_one unmaps a page just after we
370 * set VM_LOCKED, __mlock_vma_pages_range will bring it back.
374 vma->vm_flags = newflags;
376 munlock_vma_pages_range(vma, start, end);
383 static int do_mlock(unsigned long start, size_t len, int on)
385 unsigned long nstart, end, tmp;
386 struct vm_area_struct * vma, * prev;
389 VM_BUG_ON(start & ~PAGE_MASK);
390 VM_BUG_ON(len != PAGE_ALIGN(len));
396 vma = find_vma_prev(current->mm, start, &prev);
397 if (!vma || vma->vm_start > start)
400 if (start > vma->vm_start)
403 for (nstart = start ; ; ) {
404 unsigned int newflags;
406 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
408 newflags = vma->vm_flags | VM_LOCKED;
410 newflags &= ~VM_LOCKED;
415 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
419 if (nstart < prev->vm_end)
420 nstart = prev->vm_end;
425 if (!vma || vma->vm_start != nstart) {
433 static int do_mlock_pages(unsigned long start, size_t len, int ignore_errors)
435 struct mm_struct *mm = current->mm;
436 unsigned long end, nstart, nend;
437 struct vm_area_struct *vma = NULL;
441 VM_BUG_ON(start & ~PAGE_MASK);
442 VM_BUG_ON(len != PAGE_ALIGN(len));
445 for (nstart = start; nstart < end; nstart = nend) {
447 * We want to fault in pages for [nstart; end) address range.
448 * Find first corresponding VMA.
452 down_read(&mm->mmap_sem);
453 vma = find_vma(mm, nstart);
454 } else if (nstart >= vma->vm_end)
456 if (!vma || vma->vm_start >= end)
459 * Set [nstart; nend) to intersection of desired address
460 * range with the first VMA. Also, skip undesirable VMA types.
462 nend = min(end, vma->vm_end);
463 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
465 if (nstart < vma->vm_start)
466 nstart = vma->vm_start;
468 * Now fault in a range of pages. __mlock_vma_pages_range()
469 * double checks the vma flags, so that it won't mlock pages
470 * if the vma was already munlocked.
472 ret = __mlock_vma_pages_range(vma, nstart, nend, &locked);
476 continue; /* continue at next VMA */
478 ret = __mlock_posix_error_return(ret);
481 nend = nstart + ret * PAGE_SIZE;
485 up_read(&mm->mmap_sem);
486 return ret; /* 0 or negative error code */
489 SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
491 unsigned long locked;
492 unsigned long lock_limit;
498 lru_add_drain_all(); /* flush pagevec */
500 down_write(¤t->mm->mmap_sem);
501 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
504 locked = len >> PAGE_SHIFT;
505 locked += current->mm->locked_vm;
507 lock_limit = rlimit(RLIMIT_MEMLOCK);
508 lock_limit >>= PAGE_SHIFT;
510 /* check against resource limits */
511 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
512 error = do_mlock(start, len, 1);
513 up_write(¤t->mm->mmap_sem);
515 error = do_mlock_pages(start, len, 0);
519 SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
523 down_write(¤t->mm->mmap_sem);
524 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
526 ret = do_mlock(start, len, 0);
527 up_write(¤t->mm->mmap_sem);
531 static int do_mlockall(int flags)
533 struct vm_area_struct * vma, * prev = NULL;
534 unsigned int def_flags = 0;
536 if (flags & MCL_FUTURE)
537 def_flags = VM_LOCKED;
538 current->mm->def_flags = def_flags;
539 if (flags == MCL_FUTURE)
542 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
543 unsigned int newflags;
545 newflags = vma->vm_flags | VM_LOCKED;
546 if (!(flags & MCL_CURRENT))
547 newflags &= ~VM_LOCKED;
550 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
556 SYSCALL_DEFINE1(mlockall, int, flags)
558 unsigned long lock_limit;
561 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
568 lru_add_drain_all(); /* flush pagevec */
570 down_write(¤t->mm->mmap_sem);
572 lock_limit = rlimit(RLIMIT_MEMLOCK);
573 lock_limit >>= PAGE_SHIFT;
576 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
577 capable(CAP_IPC_LOCK))
578 ret = do_mlockall(flags);
579 up_write(¤t->mm->mmap_sem);
580 if (!ret && (flags & MCL_CURRENT)) {
582 do_mlock_pages(0, TASK_SIZE, 1);
588 SYSCALL_DEFINE0(munlockall)
592 down_write(¤t->mm->mmap_sem);
593 ret = do_mlockall(0);
594 up_write(¤t->mm->mmap_sem);
599 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
600 * shm segments) get accounted against the user_struct instead.
602 static DEFINE_SPINLOCK(shmlock_user_lock);
604 int user_shm_lock(size_t size, struct user_struct *user)
606 unsigned long lock_limit, locked;
609 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
610 lock_limit = rlimit(RLIMIT_MEMLOCK);
611 if (lock_limit == RLIM_INFINITY)
613 lock_limit >>= PAGE_SHIFT;
614 spin_lock(&shmlock_user_lock);
616 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
619 user->locked_shm += locked;
622 spin_unlock(&shmlock_user_lock);
626 void user_shm_unlock(size_t size, struct user_struct *user)
628 spin_lock(&shmlock_user_lock);
629 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
630 spin_unlock(&shmlock_user_lock);