thinkpad-acpi: make volume subdriver optional
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / mm / mlock.c
blob64dca47c5cbef2558a7db958f4eaa17545eb8d24
1 /*
2 * linux/mm/mlock.c
4 * (C) Copyright 1995 Linus Torvalds
5 * (C) Copyright 2002 Christoph Hellwig
6 */
8 #include <linux/capability.h>
9 #include <linux/mman.h>
10 #include <linux/mm.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>
22 #include "internal.h"
24 int can_do_mlock(void)
26 if (capable(CAP_IPC_LOCK))
27 return 1;
28 if (current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur != 0)
29 return 1;
30 return 0;
32 EXPORT_SYMBOL(can_do_mlock);
34 #ifdef CONFIG_UNEVICTABLE_LRU
36 * Mlocked pages are marked with PageMlocked() flag for efficient testing
37 * in vmscan and, possibly, the fault path; and to support semi-accurate
38 * statistics.
40 * An mlocked page [PageMlocked(page)] is unevictable. As such, it will
41 * be placed on the LRU "unevictable" list, rather than the [in]active lists.
42 * The unevictable list is an LRU sibling list to the [in]active lists.
43 * PageUnevictable is set to indicate the unevictable state.
45 * When lazy mlocking via vmscan, it is important to ensure that the
46 * vma's VM_LOCKED status is not concurrently being modified, otherwise we
47 * may have mlocked a page that is being munlocked. So lazy mlock must take
48 * the mmap_sem for read, and verify that the vma really is locked
49 * (see mm/rmap.c).
53 * LRU accounting for clear_page_mlock()
55 void __clear_page_mlock(struct page *page)
57 VM_BUG_ON(!PageLocked(page));
59 if (!page->mapping) { /* truncated ? */
60 return;
63 dec_zone_page_state(page, NR_MLOCK);
64 count_vm_event(UNEVICTABLE_PGCLEARED);
65 if (!isolate_lru_page(page)) {
66 putback_lru_page(page);
67 } else {
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 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 * called from munlock()/munmap() path with page supposedly on the LRU.
95 * Note: unlike mlock_vma_page(), we can't just clear the PageMlocked
96 * [in try_to_munlock()] and then attempt to isolate the page. We must
97 * isolate the page to keep others from messing with its unevictable
98 * and mlocked state while trying to munlock. However, we pre-clear the
99 * mlocked state anyway as we might lose the isolation race and we might
100 * not get another chance to clear PageMlocked. If we successfully
101 * isolate the page and try_to_munlock() detects other VM_LOCKED vmas
102 * mapping the page, it will restore the PageMlocked state, unless the page
103 * is mapped in a non-linear vma. So, we go ahead and SetPageMlocked(),
104 * perhaps redundantly.
105 * If we lose the isolation race, and the page is mapped by other VM_LOCKED
106 * vmas, we'll detect this in vmscan--via try_to_munlock() or try_to_unmap()
107 * either of which will restore the PageMlocked state by calling
108 * mlock_vma_page() above, if it can grab the vma's mmap sem.
110 static void munlock_vma_page(struct page *page)
112 BUG_ON(!PageLocked(page));
114 if (TestClearPageMlocked(page)) {
115 dec_zone_page_state(page, NR_MLOCK);
116 if (!isolate_lru_page(page)) {
117 int ret = try_to_munlock(page);
119 * did try_to_unlock() succeed or punt?
121 if (ret == SWAP_SUCCESS || ret == SWAP_AGAIN)
122 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
124 putback_lru_page(page);
125 } else {
127 * We lost the race. let try_to_unmap() deal
128 * with it. At least we get the page state and
129 * mlock stats right. However, page is still on
130 * the noreclaim list. We'll fix that up when
131 * the page is eventually freed or we scan the
132 * noreclaim list.
134 if (PageUnevictable(page))
135 count_vm_event(UNEVICTABLE_PGSTRANDED);
136 else
137 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
143 * __mlock_vma_pages_range() - mlock/munlock a range of pages in the vma.
144 * @vma: target vma
145 * @start: start address
146 * @end: end address
147 * @mlock: 0 indicate munlock, otherwise mlock.
149 * If @mlock == 0, unlock an mlocked range;
150 * else mlock the range of pages. This takes care of making the pages present ,
151 * 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,
159 int mlock)
161 struct mm_struct *mm = vma->vm_mm;
162 unsigned long addr = start;
163 struct page *pages[16]; /* 16 gives a reasonable batch */
164 int nr_pages = (end - start) / PAGE_SIZE;
165 int ret = 0;
166 int gup_flags = 0;
168 VM_BUG_ON(start & ~PAGE_MASK);
169 VM_BUG_ON(end & ~PAGE_MASK);
170 VM_BUG_ON(start < vma->vm_start);
171 VM_BUG_ON(end > vma->vm_end);
172 VM_BUG_ON((!rwsem_is_locked(&mm->mmap_sem)) &&
173 (atomic_read(&mm->mm_users) != 0));
176 * mlock: don't page populate if page has PROT_NONE permission.
177 * munlock: the pages always do munlock althrough
178 * its has PROT_NONE permission.
180 if (!mlock)
181 gup_flags |= GUP_FLAGS_IGNORE_VMA_PERMISSIONS;
183 if (vma->vm_flags & VM_WRITE)
184 gup_flags |= GUP_FLAGS_WRITE;
186 while (nr_pages > 0) {
187 int i;
189 cond_resched();
192 * get_user_pages makes pages present if we are
193 * setting mlock. and this extra reference count will
194 * disable migration of this page. However, page may
195 * still be truncated out from under us.
197 ret = __get_user_pages(current, mm, addr,
198 min_t(int, nr_pages, ARRAY_SIZE(pages)),
199 gup_flags, pages, NULL);
201 * This can happen for, e.g., VM_NONLINEAR regions before
202 * a page has been allocated and mapped at a given offset,
203 * or for addresses that map beyond end of a file.
204 * We'll mlock the the pages if/when they get faulted in.
206 if (ret < 0)
207 break;
208 if (ret == 0) {
210 * We know the vma is there, so the only time
211 * we cannot get a single page should be an
212 * error (ret < 0) case.
214 WARN_ON(1);
215 break;
218 lru_add_drain(); /* push cached pages to LRU */
220 for (i = 0; i < ret; i++) {
221 struct page *page = pages[i];
223 lock_page(page);
225 * Because we lock page here and migration is blocked
226 * by the elevated reference, we need only check for
227 * page truncation (file-cache only).
229 if (page->mapping) {
230 if (mlock)
231 mlock_vma_page(page);
232 else
233 munlock_vma_page(page);
235 unlock_page(page);
236 put_page(page); /* ref from get_user_pages() */
239 * here we assume that get_user_pages() has given us
240 * a list of virtually contiguous pages.
242 addr += PAGE_SIZE; /* for next get_user_pages() */
243 nr_pages--;
245 ret = 0;
248 return ret; /* count entire vma as locked_vm */
252 * convert get_user_pages() return value to posix mlock() error
254 static int __mlock_posix_error_return(long retval)
256 if (retval == -EFAULT)
257 retval = -ENOMEM;
258 else if (retval == -ENOMEM)
259 retval = -EAGAIN;
260 return retval;
263 #else /* CONFIG_UNEVICTABLE_LRU */
266 * Just make pages present if VM_LOCKED. No-op if unlocking.
268 static long __mlock_vma_pages_range(struct vm_area_struct *vma,
269 unsigned long start, unsigned long end,
270 int mlock)
272 if (mlock && (vma->vm_flags & VM_LOCKED))
273 return make_pages_present(start, end);
274 return 0;
277 static inline int __mlock_posix_error_return(long retval)
279 return 0;
282 #endif /* CONFIG_UNEVICTABLE_LRU */
285 * mlock_vma_pages_range() - mlock pages in specified vma range.
286 * @vma - the vma containing the specfied address range
287 * @start - starting address in @vma to mlock
288 * @end - end address [+1] in @vma to mlock
290 * For mmap()/mremap()/expansion of mlocked vma.
292 * return 0 on success for "normal" vmas.
294 * return number of pages [> 0] to be removed from locked_vm on success
295 * of "special" vmas.
297 long mlock_vma_pages_range(struct vm_area_struct *vma,
298 unsigned long start, unsigned long end)
300 int nr_pages = (end - start) / PAGE_SIZE;
301 BUG_ON(!(vma->vm_flags & VM_LOCKED));
304 * filter unlockable vmas
306 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
307 goto no_mlock;
309 if (!((vma->vm_flags & (VM_DONTEXPAND | VM_RESERVED)) ||
310 is_vm_hugetlb_page(vma) ||
311 vma == get_gate_vma(current))) {
313 __mlock_vma_pages_range(vma, start, end, 1);
315 /* Hide errors from mmap() and other callers */
316 return 0;
320 * User mapped kernel pages or huge pages:
321 * make these pages present to populate the ptes, but
322 * fall thru' to reset VM_LOCKED--no need to unlock, and
323 * return nr_pages so these don't get counted against task's
324 * locked limit. huge pages are already counted against
325 * locked vm limit.
327 make_pages_present(start, end);
329 no_mlock:
330 vma->vm_flags &= ~VM_LOCKED; /* and don't come back! */
331 return nr_pages; /* error or pages NOT mlocked */
336 * munlock_vma_pages_range() - munlock all pages in the vma range.'
337 * @vma - vma containing range to be munlock()ed.
338 * @start - start address in @vma of the range
339 * @end - end of range in @vma.
341 * For mremap(), munmap() and exit().
343 * Called with @vma VM_LOCKED.
345 * Returns with VM_LOCKED cleared. Callers must be prepared to
346 * deal with this.
348 * We don't save and restore VM_LOCKED here because pages are
349 * still on lru. In unmap path, pages might be scanned by reclaim
350 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
351 * free them. This will result in freeing mlocked pages.
353 void munlock_vma_pages_range(struct vm_area_struct *vma,
354 unsigned long start, unsigned long end)
356 vma->vm_flags &= ~VM_LOCKED;
357 __mlock_vma_pages_range(vma, start, end, 0);
361 * mlock_fixup - handle mlock[all]/munlock[all] requests.
363 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
364 * munlock is a no-op. However, for some special vmas, we go ahead and
365 * populate the ptes via make_pages_present().
367 * For vmas that pass the filters, merge/split as appropriate.
369 static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
370 unsigned long start, unsigned long end, unsigned int newflags)
372 struct mm_struct *mm = vma->vm_mm;
373 pgoff_t pgoff;
374 int nr_pages;
375 int ret = 0;
376 int lock = newflags & VM_LOCKED;
378 if (newflags == vma->vm_flags ||
379 (vma->vm_flags & (VM_IO | VM_PFNMAP)))
380 goto out; /* don't set VM_LOCKED, don't count */
382 if ((vma->vm_flags & (VM_DONTEXPAND | VM_RESERVED)) ||
383 is_vm_hugetlb_page(vma) ||
384 vma == get_gate_vma(current)) {
385 if (lock)
386 make_pages_present(start, end);
387 goto out; /* don't set VM_LOCKED, don't count */
390 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
391 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
392 vma->vm_file, pgoff, vma_policy(vma));
393 if (*prev) {
394 vma = *prev;
395 goto success;
398 if (start != vma->vm_start) {
399 ret = split_vma(mm, vma, start, 1);
400 if (ret)
401 goto out;
404 if (end != vma->vm_end) {
405 ret = split_vma(mm, vma, end, 0);
406 if (ret)
407 goto out;
410 success:
412 * Keep track of amount of locked VM.
414 nr_pages = (end - start) >> PAGE_SHIFT;
415 if (!lock)
416 nr_pages = -nr_pages;
417 mm->locked_vm += nr_pages;
420 * vm_flags is protected by the mmap_sem held in write mode.
421 * It's okay if try_to_unmap_one unmaps a page just after we
422 * set VM_LOCKED, __mlock_vma_pages_range will bring it back.
424 vma->vm_flags = newflags;
426 if (lock) {
427 ret = __mlock_vma_pages_range(vma, start, end, 1);
429 if (ret > 0) {
430 mm->locked_vm -= ret;
431 ret = 0;
432 } else
433 ret = __mlock_posix_error_return(ret); /* translate if needed */
434 } else {
435 __mlock_vma_pages_range(vma, start, end, 0);
438 out:
439 *prev = vma;
440 return ret;
443 static int do_mlock(unsigned long start, size_t len, int on)
445 unsigned long nstart, end, tmp;
446 struct vm_area_struct * vma, * prev;
447 int error;
449 len = PAGE_ALIGN(len);
450 end = start + len;
451 if (end < start)
452 return -EINVAL;
453 if (end == start)
454 return 0;
455 vma = find_vma_prev(current->mm, start, &prev);
456 if (!vma || vma->vm_start > start)
457 return -ENOMEM;
459 if (start > vma->vm_start)
460 prev = vma;
462 for (nstart = start ; ; ) {
463 unsigned int newflags;
465 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
467 newflags = vma->vm_flags | VM_LOCKED;
468 if (!on)
469 newflags &= ~VM_LOCKED;
471 tmp = vma->vm_end;
472 if (tmp > end)
473 tmp = end;
474 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
475 if (error)
476 break;
477 nstart = tmp;
478 if (nstart < prev->vm_end)
479 nstart = prev->vm_end;
480 if (nstart >= end)
481 break;
483 vma = prev->vm_next;
484 if (!vma || vma->vm_start != nstart) {
485 error = -ENOMEM;
486 break;
489 return error;
492 SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
494 unsigned long locked;
495 unsigned long lock_limit;
496 int error = -ENOMEM;
498 if (!can_do_mlock())
499 return -EPERM;
501 lru_add_drain_all(); /* flush pagevec */
503 down_write(&current->mm->mmap_sem);
504 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
505 start &= PAGE_MASK;
507 locked = len >> PAGE_SHIFT;
508 locked += current->mm->locked_vm;
510 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
511 lock_limit >>= PAGE_SHIFT;
513 /* check against resource limits */
514 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
515 error = do_mlock(start, len, 1);
516 up_write(&current->mm->mmap_sem);
517 return error;
520 SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
522 int ret;
524 down_write(&current->mm->mmap_sem);
525 len = PAGE_ALIGN(len + (start & ~PAGE_MASK));
526 start &= PAGE_MASK;
527 ret = do_mlock(start, len, 0);
528 up_write(&current->mm->mmap_sem);
529 return ret;
532 static int do_mlockall(int flags)
534 struct vm_area_struct * vma, * prev = NULL;
535 unsigned int def_flags = 0;
537 if (flags & MCL_FUTURE)
538 def_flags = VM_LOCKED;
539 current->mm->def_flags = def_flags;
540 if (flags == MCL_FUTURE)
541 goto out;
543 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
544 unsigned int newflags;
546 newflags = vma->vm_flags | VM_LOCKED;
547 if (!(flags & MCL_CURRENT))
548 newflags &= ~VM_LOCKED;
550 /* Ignore errors */
551 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
553 out:
554 return 0;
557 SYSCALL_DEFINE1(mlockall, int, flags)
559 unsigned long lock_limit;
560 int ret = -EINVAL;
562 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE)))
563 goto out;
565 ret = -EPERM;
566 if (!can_do_mlock())
567 goto out;
569 lru_add_drain_all(); /* flush pagevec */
571 down_write(&current->mm->mmap_sem);
573 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
574 lock_limit >>= PAGE_SHIFT;
576 ret = -ENOMEM;
577 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
578 capable(CAP_IPC_LOCK))
579 ret = do_mlockall(flags);
580 up_write(&current->mm->mmap_sem);
581 out:
582 return ret;
585 SYSCALL_DEFINE0(munlockall)
587 int ret;
589 down_write(&current->mm->mmap_sem);
590 ret = do_mlockall(0);
591 up_write(&current->mm->mmap_sem);
592 return ret;
596 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
597 * shm segments) get accounted against the user_struct instead.
599 static DEFINE_SPINLOCK(shmlock_user_lock);
601 int user_shm_lock(size_t size, struct user_struct *user)
603 unsigned long lock_limit, locked;
604 int allowed = 0;
606 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
607 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
608 if (lock_limit == RLIM_INFINITY)
609 allowed = 1;
610 lock_limit >>= PAGE_SHIFT;
611 spin_lock(&shmlock_user_lock);
612 if (!allowed &&
613 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
614 goto out;
615 get_uid(user);
616 user->locked_shm += locked;
617 allowed = 1;
618 out:
619 spin_unlock(&shmlock_user_lock);
620 return allowed;
623 void user_shm_unlock(size_t size, struct user_struct *user)
625 spin_lock(&shmlock_user_lock);
626 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
627 spin_unlock(&shmlock_user_lock);
628 free_uid(user);