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);
139 * __mlock_vma_pages_range() - mlock a range of pages in the vma.
141 * @start: start address
144 * This takes care of making the pages present too.
146 * return 0 on success, negative error code on error.
148 * vma->vm_mm->mmap_sem must be held for at least read.
150 static long __mlock_vma_pages_range(struct vm_area_struct *vma,
151 unsigned long start, unsigned long end,
154 struct mm_struct *mm = vma->vm_mm;
155 unsigned long addr = start;
156 int nr_pages = (end - start) / PAGE_SIZE;
159 VM_BUG_ON(start & ~PAGE_MASK);
160 VM_BUG_ON(end & ~PAGE_MASK);
161 VM_BUG_ON(start < vma->vm_start);
162 VM_BUG_ON(end > vma->vm_end);
163 VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem));
165 gup_flags = FOLL_TOUCH | FOLL_MLOCK;
167 * We want to touch writable mappings with a write fault in order
168 * to break COW, except for shared mappings because these don't COW
169 * and we would not want to dirty them for nothing.
171 if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
172 gup_flags |= FOLL_WRITE;
175 * We want mlock to succeed for regions that have any permissions
176 * other than PROT_NONE.
178 if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))
179 gup_flags |= FOLL_FORCE;
181 return __get_user_pages(current, mm, addr, nr_pages, gup_flags,
182 NULL, NULL, nonblocking);
186 * convert get_user_pages() return value to posix mlock() error
188 static int __mlock_posix_error_return(long retval)
190 if (retval == -EFAULT)
192 else if (retval == -ENOMEM)
198 * mlock_vma_pages_range() - mlock pages in specified vma range.
199 * @vma - the vma containing the specfied address range
200 * @start - starting address in @vma to mlock
201 * @end - end address [+1] in @vma to mlock
203 * For mmap()/mremap()/expansion of mlocked vma.
205 * return 0 on success for "normal" vmas.
207 * return number of pages [> 0] to be removed from locked_vm on success
210 long mlock_vma_pages_range(struct vm_area_struct *vma,
211 unsigned long start, unsigned long end)
213 int nr_pages = (end - start) / PAGE_SIZE;
214 BUG_ON(!(vma->vm_flags & VM_LOCKED));
217 * filter unlockable vmas
219 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
222 if (!((vma->vm_flags & (VM_DONTEXPAND | VM_RESERVED)) ||
223 is_vm_hugetlb_page(vma) ||
224 vma == get_gate_vma(current->mm))) {
226 __mlock_vma_pages_range(vma, start, end, NULL);
228 /* Hide errors from mmap() and other callers */
233 * User mapped kernel pages or huge pages:
234 * make these pages present to populate the ptes, but
235 * fall thru' to reset VM_LOCKED--no need to unlock, and
236 * return nr_pages so these don't get counted against task's
237 * locked limit. huge pages are already counted against
240 make_pages_present(start, end);
243 vma->vm_flags &= ~VM_LOCKED; /* and don't come back! */
244 return nr_pages; /* error or pages NOT mlocked */
248 * munlock_vma_pages_range() - munlock all pages in the vma range.'
249 * @vma - vma containing range to be munlock()ed.
250 * @start - start address in @vma of the range
251 * @end - end of range in @vma.
253 * For mremap(), munmap() and exit().
255 * Called with @vma VM_LOCKED.
257 * Returns with VM_LOCKED cleared. Callers must be prepared to
260 * We don't save and restore VM_LOCKED here because pages are
261 * still on lru. In unmap path, pages might be scanned by reclaim
262 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
263 * free them. This will result in freeing mlocked pages.
265 void munlock_vma_pages_range(struct vm_area_struct *vma,
266 unsigned long start, unsigned long end)
271 vma->vm_flags &= ~VM_LOCKED;
273 for (addr = start; addr < end; addr += PAGE_SIZE) {
276 * Although FOLL_DUMP is intended for get_dump_page(),
277 * it just so happens that its special treatment of the
278 * ZERO_PAGE (returning an error instead of doing get_page)
279 * suits munlock very well (and if somehow an abnormal page
280 * has sneaked into the range, we won't oops here: great).
282 page = follow_page(vma, addr, FOLL_GET | FOLL_DUMP);
283 if (page && !IS_ERR(page)) {
286 * Like in __mlock_vma_pages_range(),
287 * because we lock page here and migration is
288 * blocked by the elevated reference, we need
289 * only check for file-cache page truncation.
292 munlock_vma_page(page);
301 * mlock_fixup - handle mlock[all]/munlock[all] requests.
303 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
304 * munlock is a no-op. However, for some special vmas, we go ahead and
305 * populate the ptes via make_pages_present().
307 * For vmas that pass the filters, merge/split as appropriate.
309 static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
310 unsigned long start, unsigned long end, vm_flags_t newflags)
312 struct mm_struct *mm = vma->vm_mm;
316 int lock = !!(newflags & VM_LOCKED);
318 if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
319 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
320 goto out; /* don't set VM_LOCKED, don't count */
322 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
323 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
324 vma->vm_file, pgoff, vma_policy(vma));
330 if (start != vma->vm_start) {
331 ret = split_vma(mm, vma, start, 1);
336 if (end != vma->vm_end) {
337 ret = split_vma(mm, vma, end, 0);
344 * Keep track of amount of locked VM.
346 nr_pages = (end - start) >> PAGE_SHIFT;
348 nr_pages = -nr_pages;
349 mm->locked_vm += nr_pages;
352 * vm_flags is protected by the mmap_sem held in write mode.
353 * It's okay if try_to_unmap_one unmaps a page just after we
354 * set VM_LOCKED, __mlock_vma_pages_range will bring it back.
358 vma->vm_flags = newflags;
360 munlock_vma_pages_range(vma, start, end);
367 static int do_mlock(unsigned long start, size_t len, int on)
369 unsigned long nstart, end, tmp;
370 struct vm_area_struct * vma, * prev;
373 VM_BUG_ON(start & ~PAGE_MASK);
374 VM_BUG_ON(len != PAGE_ALIGN(len));
380 vma = find_vma_prev(current->mm, start, &prev);
381 if (!vma || vma->vm_start > start)
384 if (start > vma->vm_start)
387 for (nstart = start ; ; ) {
390 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
392 newflags = vma->vm_flags | VM_LOCKED;
394 newflags &= ~VM_LOCKED;
399 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
403 if (nstart < prev->vm_end)
404 nstart = prev->vm_end;
409 if (!vma || vma->vm_start != nstart) {
417 static int do_mlock_pages(unsigned long start, size_t len, int ignore_errors)
419 struct mm_struct *mm = current->mm;
420 unsigned long end, nstart, nend;
421 struct vm_area_struct *vma = NULL;
425 VM_BUG_ON(start & ~PAGE_MASK);
426 VM_BUG_ON(len != PAGE_ALIGN(len));
429 for (nstart = start; nstart < end; nstart = nend) {
431 * We want to fault in pages for [nstart; end) address range.
432 * Find first corresponding VMA.
436 down_read(&mm->mmap_sem);
437 vma = find_vma(mm, nstart);
438 } else if (nstart >= vma->vm_end)
440 if (!vma || vma->vm_start >= end)
443 * Set [nstart; nend) to intersection of desired address
444 * range with the first VMA. Also, skip undesirable VMA types.
446 nend = min(end, vma->vm_end);
447 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
449 if (nstart < vma->vm_start)
450 nstart = vma->vm_start;
452 * Now fault in a range of pages. __mlock_vma_pages_range()
453 * double checks the vma flags, so that it won't mlock pages
454 * if the vma was already munlocked.
456 ret = __mlock_vma_pages_range(vma, nstart, nend, &locked);
460 continue; /* continue at next VMA */
462 ret = __mlock_posix_error_return(ret);
465 nend = nstart + ret * PAGE_SIZE;
469 up_read(&mm->mmap_sem);
470 return ret; /* 0 or negative error code */
473 SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
475 unsigned long locked;
476 unsigned long lock_limit;
482 lru_add_drain_all(); /* flush pagevec */
484 down_write(¤t->mm->mmap_sem);
485 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
488 locked = len >> PAGE_SHIFT;
489 locked += current->mm->locked_vm;
491 lock_limit = rlimit(RLIMIT_MEMLOCK);
492 lock_limit >>= PAGE_SHIFT;
494 /* check against resource limits */
495 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
496 error = do_mlock(start, len, 1);
497 up_write(¤t->mm->mmap_sem);
499 error = do_mlock_pages(start, len, 0);
503 SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
507 down_write(¤t->mm->mmap_sem);
508 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
510 ret = do_mlock(start, len, 0);
511 up_write(¤t->mm->mmap_sem);
515 static int do_mlockall(int flags)
517 struct vm_area_struct * vma, * prev = NULL;
518 unsigned int def_flags = 0;
520 if (flags & MCL_FUTURE)
521 def_flags = VM_LOCKED;
522 current->mm->def_flags = def_flags;
523 if (flags == MCL_FUTURE)
526 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
529 newflags = vma->vm_flags | VM_LOCKED;
530 if (!(flags & MCL_CURRENT))
531 newflags &= ~VM_LOCKED;
534 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
540 SYSCALL_DEFINE1(mlockall, int, flags)
542 unsigned long lock_limit;
545 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
552 lru_add_drain_all(); /* flush pagevec */
554 down_write(¤t->mm->mmap_sem);
556 lock_limit = rlimit(RLIMIT_MEMLOCK);
557 lock_limit >>= PAGE_SHIFT;
560 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
561 capable(CAP_IPC_LOCK))
562 ret = do_mlockall(flags);
563 up_write(¤t->mm->mmap_sem);
564 if (!ret && (flags & MCL_CURRENT)) {
566 do_mlock_pages(0, TASK_SIZE, 1);
572 SYSCALL_DEFINE0(munlockall)
576 down_write(¤t->mm->mmap_sem);
577 ret = do_mlockall(0);
578 up_write(¤t->mm->mmap_sem);
583 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
584 * shm segments) get accounted against the user_struct instead.
586 static DEFINE_SPINLOCK(shmlock_user_lock);
588 int user_shm_lock(size_t size, struct user_struct *user)
590 unsigned long lock_limit, locked;
593 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
594 lock_limit = rlimit(RLIMIT_MEMLOCK);
595 if (lock_limit == RLIM_INFINITY)
597 lock_limit >>= PAGE_SHIFT;
598 spin_lock(&shmlock_user_lock);
600 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
603 user->locked_shm += locked;
606 spin_unlock(&shmlock_user_lock);
610 void user_shm_unlock(size_t size, struct user_struct *user)
612 spin_lock(&shmlock_user_lock);
613 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
614 spin_unlock(&shmlock_user_lock);