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 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 /* Serialize with page migration */
82 BUG_ON(!PageLocked(page));
84 if (!TestSetPageMlocked(page)) {
85 inc_zone_page_state(page, NR_MLOCK);
86 count_vm_event(UNEVICTABLE_PGMLOCKED);
87 if (!isolate_lru_page(page))
88 putback_lru_page(page);
93 * munlock_vma_page - munlock a vma page
94 * @page - page to be unlocked
96 * called from munlock()/munmap() path with page supposedly on the LRU.
97 * When we munlock a page, because the vma where we found the page is being
98 * munlock()ed or munmap()ed, we want to check whether other vmas hold the
99 * page locked so that we can leave it on the unevictable lru list and not
100 * bother vmscan with it. However, to walk the page's rmap list in
101 * try_to_munlock() we must isolate the page from the LRU. If some other
102 * task has removed the page from the LRU, we won't be able to do that.
103 * So we clear the PageMlocked as we might not get another chance. If we
104 * can't isolate the page, we leave it for putback_lru_page() and vmscan
105 * [page_referenced()/try_to_unmap()] to deal with.
107 void munlock_vma_page(struct page *page)
109 /* For try_to_munlock() and to serialize with page migration */
110 BUG_ON(!PageLocked(page));
112 if (TestClearPageMlocked(page)) {
113 dec_zone_page_state(page, NR_MLOCK);
114 if (!isolate_lru_page(page)) {
115 int ret = SWAP_AGAIN;
118 * Optimization: if the page was mapped just once,
119 * that's our mapping and we don't need to check all the
122 if (page_mapcount(page) > 1)
123 ret = try_to_munlock(page);
125 * did try_to_unlock() succeed or punt?
127 if (ret != SWAP_MLOCK)
128 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
130 putback_lru_page(page);
133 * Some other task has removed the page from the LRU.
134 * putback_lru_page() will take care of removing the
135 * page from the unevictable list, if necessary.
136 * vmscan [page_referenced()] will move the page back
137 * to the unevictable list if some other vma has it
140 if (PageUnevictable(page))
141 count_vm_event(UNEVICTABLE_PGSTRANDED);
143 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
149 * __mlock_vma_pages_range() - mlock a range of pages in the vma.
151 * @start: start address
154 * This takes care of making the pages present too.
156 * return 0 on success, negative error code on error.
158 * vma->vm_mm->mmap_sem must be held for at least read.
160 static long __mlock_vma_pages_range(struct vm_area_struct *vma,
161 unsigned long start, unsigned long end,
164 struct mm_struct *mm = vma->vm_mm;
165 unsigned long addr = start;
166 int nr_pages = (end - start) / PAGE_SIZE;
169 VM_BUG_ON(start & ~PAGE_MASK);
170 VM_BUG_ON(end & ~PAGE_MASK);
171 VM_BUG_ON(start < vma->vm_start);
172 VM_BUG_ON(end > vma->vm_end);
173 VM_BUG_ON(!rwsem_is_locked(&mm->mmap_sem));
175 gup_flags = FOLL_TOUCH | FOLL_MLOCK;
177 * We want to touch writable mappings with a write fault in order
178 * to break COW, except for shared mappings because these don't COW
179 * and we would not want to dirty them for nothing.
181 if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
182 gup_flags |= FOLL_WRITE;
185 * We want mlock to succeed for regions that have any permissions
186 * other than PROT_NONE.
188 if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))
189 gup_flags |= FOLL_FORCE;
191 return __get_user_pages(current, mm, addr, nr_pages, gup_flags,
192 NULL, NULL, nonblocking);
196 * convert get_user_pages() return value to posix mlock() error
198 static int __mlock_posix_error_return(long retval)
200 if (retval == -EFAULT)
202 else if (retval == -ENOMEM)
208 * mlock_vma_pages_range() - mlock pages in specified vma range.
209 * @vma - the vma containing the specfied address range
210 * @start - starting address in @vma to mlock
211 * @end - end address [+1] in @vma to mlock
213 * For mmap()/mremap()/expansion of mlocked vma.
215 * return 0 on success for "normal" vmas.
217 * return number of pages [> 0] to be removed from locked_vm on success
220 long mlock_vma_pages_range(struct vm_area_struct *vma,
221 unsigned long start, unsigned long end)
223 int nr_pages = (end - start) / PAGE_SIZE;
224 BUG_ON(!(vma->vm_flags & VM_LOCKED));
227 * filter unlockable vmas
229 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
232 if (!((vma->vm_flags & (VM_DONTEXPAND | VM_RESERVED)) ||
233 is_vm_hugetlb_page(vma) ||
234 vma == get_gate_vma(current->mm))) {
236 __mlock_vma_pages_range(vma, start, end, NULL);
238 /* Hide errors from mmap() and other callers */
243 * User mapped kernel pages or huge pages:
244 * make these pages present to populate the ptes, but
245 * fall thru' to reset VM_LOCKED--no need to unlock, and
246 * return nr_pages so these don't get counted against task's
247 * locked limit. huge pages are already counted against
250 make_pages_present(start, end);
253 vma->vm_flags &= ~VM_LOCKED; /* and don't come back! */
254 return nr_pages; /* error or pages NOT mlocked */
258 * munlock_vma_pages_range() - munlock all pages in the vma range.'
259 * @vma - vma containing range to be munlock()ed.
260 * @start - start address in @vma of the range
261 * @end - end of range in @vma.
263 * For mremap(), munmap() and exit().
265 * Called with @vma VM_LOCKED.
267 * Returns with VM_LOCKED cleared. Callers must be prepared to
270 * We don't save and restore VM_LOCKED here because pages are
271 * still on lru. In unmap path, pages might be scanned by reclaim
272 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
273 * free them. This will result in freeing mlocked pages.
275 void munlock_vma_pages_range(struct vm_area_struct *vma,
276 unsigned long start, unsigned long end)
281 vma->vm_flags &= ~VM_LOCKED;
283 for (addr = start; addr < end; addr += PAGE_SIZE) {
286 * Although FOLL_DUMP is intended for get_dump_page(),
287 * it just so happens that its special treatment of the
288 * ZERO_PAGE (returning an error instead of doing get_page)
289 * suits munlock very well (and if somehow an abnormal page
290 * has sneaked into the range, we won't oops here: great).
292 page = follow_page(vma, addr, FOLL_GET | FOLL_DUMP);
293 if (page && !IS_ERR(page)) {
296 * Like in __mlock_vma_pages_range(),
297 * because we lock page here and migration is
298 * blocked by the elevated reference, we need
299 * only check for file-cache page truncation.
302 munlock_vma_page(page);
311 * mlock_fixup - handle mlock[all]/munlock[all] requests.
313 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
314 * munlock is a no-op. However, for some special vmas, we go ahead and
315 * populate the ptes via make_pages_present().
317 * For vmas that pass the filters, merge/split as appropriate.
319 static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
320 unsigned long start, unsigned long end, vm_flags_t newflags)
322 struct mm_struct *mm = vma->vm_mm;
326 int lock = !!(newflags & VM_LOCKED);
328 if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
329 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
330 goto out; /* don't set VM_LOCKED, don't count */
332 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
333 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
334 vma->vm_file, pgoff, vma_policy(vma));
340 if (start != vma->vm_start) {
341 ret = split_vma(mm, vma, start, 1);
346 if (end != vma->vm_end) {
347 ret = split_vma(mm, vma, end, 0);
354 * Keep track of amount of locked VM.
356 nr_pages = (end - start) >> PAGE_SHIFT;
358 nr_pages = -nr_pages;
359 mm->locked_vm += nr_pages;
362 * vm_flags is protected by the mmap_sem held in write mode.
363 * It's okay if try_to_unmap_one unmaps a page just after we
364 * set VM_LOCKED, __mlock_vma_pages_range will bring it back.
368 vma->vm_flags = newflags;
370 munlock_vma_pages_range(vma, start, end);
377 static int do_mlock(unsigned long start, size_t len, int on)
379 unsigned long nstart, end, tmp;
380 struct vm_area_struct * vma, * prev;
383 VM_BUG_ON(start & ~PAGE_MASK);
384 VM_BUG_ON(len != PAGE_ALIGN(len));
390 vma = find_vma_prev(current->mm, start, &prev);
391 if (!vma || vma->vm_start > start)
394 if (start > vma->vm_start)
397 for (nstart = start ; ; ) {
400 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
402 newflags = vma->vm_flags | VM_LOCKED;
404 newflags &= ~VM_LOCKED;
409 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
413 if (nstart < prev->vm_end)
414 nstart = prev->vm_end;
419 if (!vma || vma->vm_start != nstart) {
427 static int do_mlock_pages(unsigned long start, size_t len, int ignore_errors)
429 struct mm_struct *mm = current->mm;
430 unsigned long end, nstart, nend;
431 struct vm_area_struct *vma = NULL;
435 VM_BUG_ON(start & ~PAGE_MASK);
436 VM_BUG_ON(len != PAGE_ALIGN(len));
439 for (nstart = start; nstart < end; nstart = nend) {
441 * We want to fault in pages for [nstart; end) address range.
442 * Find first corresponding VMA.
446 down_read(&mm->mmap_sem);
447 vma = find_vma(mm, nstart);
448 } else if (nstart >= vma->vm_end)
450 if (!vma || vma->vm_start >= end)
453 * Set [nstart; nend) to intersection of desired address
454 * range with the first VMA. Also, skip undesirable VMA types.
456 nend = min(end, vma->vm_end);
457 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
459 if (nstart < vma->vm_start)
460 nstart = vma->vm_start;
462 * Now fault in a range of pages. __mlock_vma_pages_range()
463 * double checks the vma flags, so that it won't mlock pages
464 * if the vma was already munlocked.
466 ret = __mlock_vma_pages_range(vma, nstart, nend, &locked);
470 continue; /* continue at next VMA */
472 ret = __mlock_posix_error_return(ret);
475 nend = nstart + ret * PAGE_SIZE;
479 up_read(&mm->mmap_sem);
480 return ret; /* 0 or negative error code */
483 SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
485 unsigned long locked;
486 unsigned long lock_limit;
492 lru_add_drain_all(); /* flush pagevec */
494 down_write(¤t->mm->mmap_sem);
495 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
498 locked = len >> PAGE_SHIFT;
499 locked += current->mm->locked_vm;
501 lock_limit = rlimit(RLIMIT_MEMLOCK);
502 lock_limit >>= PAGE_SHIFT;
504 /* check against resource limits */
505 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
506 error = do_mlock(start, len, 1);
507 up_write(¤t->mm->mmap_sem);
509 error = do_mlock_pages(start, len, 0);
513 SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
517 down_write(¤t->mm->mmap_sem);
518 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
520 ret = do_mlock(start, len, 0);
521 up_write(¤t->mm->mmap_sem);
525 static int do_mlockall(int flags)
527 struct vm_area_struct * vma, * prev = NULL;
528 unsigned int def_flags = 0;
530 if (flags & MCL_FUTURE)
531 def_flags = VM_LOCKED;
532 current->mm->def_flags = def_flags;
533 if (flags == MCL_FUTURE)
536 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
539 newflags = vma->vm_flags | VM_LOCKED;
540 if (!(flags & MCL_CURRENT))
541 newflags &= ~VM_LOCKED;
544 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
550 SYSCALL_DEFINE1(mlockall, int, flags)
552 unsigned long lock_limit;
555 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
562 if (flags & MCL_CURRENT)
563 lru_add_drain_all(); /* flush pagevec */
565 down_write(¤t->mm->mmap_sem);
567 lock_limit = rlimit(RLIMIT_MEMLOCK);
568 lock_limit >>= PAGE_SHIFT;
571 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
572 capable(CAP_IPC_LOCK))
573 ret = do_mlockall(flags);
574 up_write(¤t->mm->mmap_sem);
575 if (!ret && (flags & MCL_CURRENT)) {
577 do_mlock_pages(0, TASK_SIZE, 1);
583 SYSCALL_DEFINE0(munlockall)
587 down_write(¤t->mm->mmap_sem);
588 ret = do_mlockall(0);
589 up_write(¤t->mm->mmap_sem);
594 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
595 * shm segments) get accounted against the user_struct instead.
597 static DEFINE_SPINLOCK(shmlock_user_lock);
599 int user_shm_lock(size_t size, struct user_struct *user)
601 unsigned long lock_limit, locked;
604 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
605 lock_limit = rlimit(RLIMIT_MEMLOCK);
606 if (lock_limit == RLIM_INFINITY)
608 lock_limit >>= PAGE_SHIFT;
609 spin_lock(&shmlock_user_lock);
611 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
614 user->locked_shm += locked;
617 spin_unlock(&shmlock_user_lock);
621 void user_shm_unlock(size_t size, struct user_struct *user)
623 spin_lock(&shmlock_user_lock);
624 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
625 spin_unlock(&shmlock_user_lock);