Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/mm/mlock.c
4 *
5 * (C) Copyright 1995 Linus Torvalds
6 * (C) Copyright 2002 Christoph Hellwig
7 */
8
9#include <linux/capability.h>
10#include <linux/mman.h>
11#include <linux/mm.h>
12#include <linux/sched/user.h>
13#include <linux/swap.h>
14#include <linux/swapops.h>
15#include <linux/pagemap.h>
16#include <linux/pagevec.h>
17#include <linux/pagewalk.h>
18#include <linux/mempolicy.h>
19#include <linux/syscalls.h>
20#include <linux/sched.h>
21#include <linux/export.h>
22#include <linux/rmap.h>
23#include <linux/mmzone.h>
24#include <linux/hugetlb.h>
25#include <linux/memcontrol.h>
26#include <linux/mm_inline.h>
27#include <linux/secretmem.h>
28
29#include "internal.h"
30
31struct mlock_fbatch {
32 local_lock_t lock;
33 struct folio_batch fbatch;
34};
35
36static DEFINE_PER_CPU(struct mlock_fbatch, mlock_fbatch) = {
37 .lock = INIT_LOCAL_LOCK(lock),
38};
39
40bool can_do_mlock(void)
41{
42 if (rlimit(RLIMIT_MEMLOCK) != 0)
43 return true;
44 if (capable(CAP_IPC_LOCK))
45 return true;
46 return false;
47}
48EXPORT_SYMBOL(can_do_mlock);
49
50/*
51 * Mlocked folios are marked with the PG_mlocked flag for efficient testing
52 * in vmscan and, possibly, the fault path; and to support semi-accurate
53 * statistics.
54 *
55 * An mlocked folio [folio_test_mlocked(folio)] is unevictable. As such, it
56 * will be ostensibly placed on the LRU "unevictable" list (actually no such
57 * list exists), rather than the [in]active lists. PG_unevictable is set to
58 * indicate the unevictable state.
59 */
60
61static struct lruvec *__mlock_folio(struct folio *folio, struct lruvec *lruvec)
62{
63 /* There is nothing more we can do while it's off LRU */
64 if (!folio_test_clear_lru(folio))
65 return lruvec;
66
67 lruvec = folio_lruvec_relock_irq(folio, lruvec);
68
69 if (unlikely(folio_evictable(folio))) {
70 /*
71 * This is a little surprising, but quite possible: PG_mlocked
72 * must have got cleared already by another CPU. Could this
73 * folio be unevictable? I'm not sure, but move it now if so.
74 */
75 if (folio_test_unevictable(folio)) {
76 lruvec_del_folio(lruvec, folio);
77 folio_clear_unevictable(folio);
78 lruvec_add_folio(lruvec, folio);
79
80 __count_vm_events(UNEVICTABLE_PGRESCUED,
81 folio_nr_pages(folio));
82 }
83 goto out;
84 }
85
86 if (folio_test_unevictable(folio)) {
87 if (folio_test_mlocked(folio))
88 folio->mlock_count++;
89 goto out;
90 }
91
92 lruvec_del_folio(lruvec, folio);
93 folio_clear_active(folio);
94 folio_set_unevictable(folio);
95 folio->mlock_count = !!folio_test_mlocked(folio);
96 lruvec_add_folio(lruvec, folio);
97 __count_vm_events(UNEVICTABLE_PGCULLED, folio_nr_pages(folio));
98out:
99 folio_set_lru(folio);
100 return lruvec;
101}
102
103static struct lruvec *__mlock_new_folio(struct folio *folio, struct lruvec *lruvec)
104{
105 VM_BUG_ON_FOLIO(folio_test_lru(folio), folio);
106
107 lruvec = folio_lruvec_relock_irq(folio, lruvec);
108
109 /* As above, this is a little surprising, but possible */
110 if (unlikely(folio_evictable(folio)))
111 goto out;
112
113 folio_set_unevictable(folio);
114 folio->mlock_count = !!folio_test_mlocked(folio);
115 __count_vm_events(UNEVICTABLE_PGCULLED, folio_nr_pages(folio));
116out:
117 lruvec_add_folio(lruvec, folio);
118 folio_set_lru(folio);
119 return lruvec;
120}
121
122static struct lruvec *__munlock_folio(struct folio *folio, struct lruvec *lruvec)
123{
124 int nr_pages = folio_nr_pages(folio);
125 bool isolated = false;
126
127 if (!folio_test_clear_lru(folio))
128 goto munlock;
129
130 isolated = true;
131 lruvec = folio_lruvec_relock_irq(folio, lruvec);
132
133 if (folio_test_unevictable(folio)) {
134 /* Then mlock_count is maintained, but might undercount */
135 if (folio->mlock_count)
136 folio->mlock_count--;
137 if (folio->mlock_count)
138 goto out;
139 }
140 /* else assume that was the last mlock: reclaim will fix it if not */
141
142munlock:
143 if (folio_test_clear_mlocked(folio)) {
144 __zone_stat_mod_folio(folio, NR_MLOCK, -nr_pages);
145 if (isolated || !folio_test_unevictable(folio))
146 __count_vm_events(UNEVICTABLE_PGMUNLOCKED, nr_pages);
147 else
148 __count_vm_events(UNEVICTABLE_PGSTRANDED, nr_pages);
149 }
150
151 /* folio_evictable() has to be checked *after* clearing Mlocked */
152 if (isolated && folio_test_unevictable(folio) && folio_evictable(folio)) {
153 lruvec_del_folio(lruvec, folio);
154 folio_clear_unevictable(folio);
155 lruvec_add_folio(lruvec, folio);
156 __count_vm_events(UNEVICTABLE_PGRESCUED, nr_pages);
157 }
158out:
159 if (isolated)
160 folio_set_lru(folio);
161 return lruvec;
162}
163
164/*
165 * Flags held in the low bits of a struct folio pointer on the mlock_fbatch.
166 */
167#define LRU_FOLIO 0x1
168#define NEW_FOLIO 0x2
169static inline struct folio *mlock_lru(struct folio *folio)
170{
171 return (struct folio *)((unsigned long)folio + LRU_FOLIO);
172}
173
174static inline struct folio *mlock_new(struct folio *folio)
175{
176 return (struct folio *)((unsigned long)folio + NEW_FOLIO);
177}
178
179/*
180 * mlock_folio_batch() is derived from folio_batch_move_lru(): perhaps that can
181 * make use of such folio pointer flags in future, but for now just keep it for
182 * mlock. We could use three separate folio batches instead, but one feels
183 * better (munlocking a full folio batch does not need to drain mlocking folio
184 * batches first).
185 */
186static void mlock_folio_batch(struct folio_batch *fbatch)
187{
188 struct lruvec *lruvec = NULL;
189 unsigned long mlock;
190 struct folio *folio;
191 int i;
192
193 for (i = 0; i < folio_batch_count(fbatch); i++) {
194 folio = fbatch->folios[i];
195 mlock = (unsigned long)folio & (LRU_FOLIO | NEW_FOLIO);
196 folio = (struct folio *)((unsigned long)folio - mlock);
197 fbatch->folios[i] = folio;
198
199 if (mlock & LRU_FOLIO)
200 lruvec = __mlock_folio(folio, lruvec);
201 else if (mlock & NEW_FOLIO)
202 lruvec = __mlock_new_folio(folio, lruvec);
203 else
204 lruvec = __munlock_folio(folio, lruvec);
205 }
206
207 if (lruvec)
208 unlock_page_lruvec_irq(lruvec);
209 folios_put(fbatch->folios, folio_batch_count(fbatch));
210 folio_batch_reinit(fbatch);
211}
212
213void mlock_drain_local(void)
214{
215 struct folio_batch *fbatch;
216
217 local_lock(&mlock_fbatch.lock);
218 fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
219 if (folio_batch_count(fbatch))
220 mlock_folio_batch(fbatch);
221 local_unlock(&mlock_fbatch.lock);
222}
223
224void mlock_drain_remote(int cpu)
225{
226 struct folio_batch *fbatch;
227
228 WARN_ON_ONCE(cpu_online(cpu));
229 fbatch = &per_cpu(mlock_fbatch.fbatch, cpu);
230 if (folio_batch_count(fbatch))
231 mlock_folio_batch(fbatch);
232}
233
234bool need_mlock_drain(int cpu)
235{
236 return folio_batch_count(&per_cpu(mlock_fbatch.fbatch, cpu));
237}
238
239/**
240 * mlock_folio - mlock a folio already on (or temporarily off) LRU
241 * @folio: folio to be mlocked.
242 */
243void mlock_folio(struct folio *folio)
244{
245 struct folio_batch *fbatch;
246
247 local_lock(&mlock_fbatch.lock);
248 fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
249
250 if (!folio_test_set_mlocked(folio)) {
251 int nr_pages = folio_nr_pages(folio);
252
253 zone_stat_mod_folio(folio, NR_MLOCK, nr_pages);
254 __count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
255 }
256
257 folio_get(folio);
258 if (!folio_batch_add(fbatch, mlock_lru(folio)) ||
259 folio_test_large(folio) || lru_cache_disabled())
260 mlock_folio_batch(fbatch);
261 local_unlock(&mlock_fbatch.lock);
262}
263
264/**
265 * mlock_new_folio - mlock a newly allocated folio not yet on LRU
266 * @folio: folio to be mlocked, either normal or a THP head.
267 */
268void mlock_new_folio(struct folio *folio)
269{
270 struct folio_batch *fbatch;
271 int nr_pages = folio_nr_pages(folio);
272
273 local_lock(&mlock_fbatch.lock);
274 fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
275 folio_set_mlocked(folio);
276
277 zone_stat_mod_folio(folio, NR_MLOCK, nr_pages);
278 __count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
279
280 folio_get(folio);
281 if (!folio_batch_add(fbatch, mlock_new(folio)) ||
282 folio_test_large(folio) || lru_cache_disabled())
283 mlock_folio_batch(fbatch);
284 local_unlock(&mlock_fbatch.lock);
285}
286
287/**
288 * munlock_folio - munlock a folio
289 * @folio: folio to be munlocked, either normal or a THP head.
290 */
291void munlock_folio(struct folio *folio)
292{
293 struct folio_batch *fbatch;
294
295 local_lock(&mlock_fbatch.lock);
296 fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
297 /*
298 * folio_test_clear_mlocked(folio) must be left to __munlock_folio(),
299 * which will check whether the folio is multiply mlocked.
300 */
301 folio_get(folio);
302 if (!folio_batch_add(fbatch, folio) ||
303 folio_test_large(folio) || lru_cache_disabled())
304 mlock_folio_batch(fbatch);
305 local_unlock(&mlock_fbatch.lock);
306}
307
308static inline unsigned int folio_mlock_step(struct folio *folio,
309 pte_t *pte, unsigned long addr, unsigned long end)
310{
311 unsigned int count, i, nr = folio_nr_pages(folio);
312 unsigned long pfn = folio_pfn(folio);
313 pte_t ptent = ptep_get(pte);
314
315 if (!folio_test_large(folio))
316 return 1;
317
318 count = pfn + nr - pte_pfn(ptent);
319 count = min_t(unsigned int, count, (end - addr) >> PAGE_SHIFT);
320
321 for (i = 0; i < count; i++, pte++) {
322 pte_t entry = ptep_get(pte);
323
324 if (!pte_present(entry))
325 break;
326 if (pte_pfn(entry) - pfn >= nr)
327 break;
328 }
329
330 return i;
331}
332
333static inline bool allow_mlock_munlock(struct folio *folio,
334 struct vm_area_struct *vma, unsigned long start,
335 unsigned long end, unsigned int step)
336{
337 /*
338 * For unlock, allow munlock large folio which is partially
339 * mapped to VMA. As it's possible that large folio is
340 * mlocked and VMA is split later.
341 *
342 * During memory pressure, such kind of large folio can
343 * be split. And the pages are not in VM_LOCKed VMA
344 * can be reclaimed.
345 */
346 if (!(vma->vm_flags & VM_LOCKED))
347 return true;
348
349 /* folio_within_range() cannot take KSM, but any small folio is OK */
350 if (!folio_test_large(folio))
351 return true;
352
353 /* folio not in range [start, end), skip mlock */
354 if (!folio_within_range(folio, vma, start, end))
355 return false;
356
357 /* folio is not fully mapped, skip mlock */
358 if (step != folio_nr_pages(folio))
359 return false;
360
361 return true;
362}
363
364static int mlock_pte_range(pmd_t *pmd, unsigned long addr,
365 unsigned long end, struct mm_walk *walk)
366
367{
368 struct vm_area_struct *vma = walk->vma;
369 spinlock_t *ptl;
370 pte_t *start_pte, *pte;
371 pte_t ptent;
372 struct folio *folio;
373 unsigned int step = 1;
374 unsigned long start = addr;
375
376 ptl = pmd_trans_huge_lock(pmd, vma);
377 if (ptl) {
378 if (!pmd_present(*pmd))
379 goto out;
380 if (is_huge_zero_pmd(*pmd))
381 goto out;
382 folio = page_folio(pmd_page(*pmd));
383 if (vma->vm_flags & VM_LOCKED)
384 mlock_folio(folio);
385 else
386 munlock_folio(folio);
387 goto out;
388 }
389
390 start_pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
391 if (!start_pte) {
392 walk->action = ACTION_AGAIN;
393 return 0;
394 }
395
396 for (pte = start_pte; addr != end; pte++, addr += PAGE_SIZE) {
397 ptent = ptep_get(pte);
398 if (!pte_present(ptent))
399 continue;
400 folio = vm_normal_folio(vma, addr, ptent);
401 if (!folio || folio_is_zone_device(folio))
402 continue;
403
404 step = folio_mlock_step(folio, pte, addr, end);
405 if (!allow_mlock_munlock(folio, vma, start, end, step))
406 goto next_entry;
407
408 if (vma->vm_flags & VM_LOCKED)
409 mlock_folio(folio);
410 else
411 munlock_folio(folio);
412
413next_entry:
414 pte += step - 1;
415 addr += (step - 1) << PAGE_SHIFT;
416 }
417 pte_unmap(start_pte);
418out:
419 spin_unlock(ptl);
420 cond_resched();
421 return 0;
422}
423
424/*
425 * mlock_vma_pages_range() - mlock any pages already in the range,
426 * or munlock all pages in the range.
427 * @vma - vma containing range to be mlock()ed or munlock()ed
428 * @start - start address in @vma of the range
429 * @end - end of range in @vma
430 * @newflags - the new set of flags for @vma.
431 *
432 * Called for mlock(), mlock2() and mlockall(), to set @vma VM_LOCKED;
433 * called for munlock() and munlockall(), to clear VM_LOCKED from @vma.
434 */
435static void mlock_vma_pages_range(struct vm_area_struct *vma,
436 unsigned long start, unsigned long end, vm_flags_t newflags)
437{
438 static const struct mm_walk_ops mlock_walk_ops = {
439 .pmd_entry = mlock_pte_range,
440 .walk_lock = PGWALK_WRLOCK_VERIFY,
441 };
442
443 /*
444 * There is a slight chance that concurrent page migration,
445 * or page reclaim finding a page of this now-VM_LOCKED vma,
446 * will call mlock_vma_folio() and raise page's mlock_count:
447 * double counting, leaving the page unevictable indefinitely.
448 * Communicate this danger to mlock_vma_folio() with VM_IO,
449 * which is a VM_SPECIAL flag not allowed on VM_LOCKED vmas.
450 * mmap_lock is held in write mode here, so this weird
451 * combination should not be visible to other mmap_lock users;
452 * but WRITE_ONCE so rmap walkers must see VM_IO if VM_LOCKED.
453 */
454 if (newflags & VM_LOCKED)
455 newflags |= VM_IO;
456 vma_start_write(vma);
457 vm_flags_reset_once(vma, newflags);
458
459 lru_add_drain();
460 walk_page_range(vma->vm_mm, start, end, &mlock_walk_ops, NULL);
461 lru_add_drain();
462
463 if (newflags & VM_IO) {
464 newflags &= ~VM_IO;
465 vm_flags_reset_once(vma, newflags);
466 }
467}
468
469/*
470 * mlock_fixup - handle mlock[all]/munlock[all] requests.
471 *
472 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
473 * munlock is a no-op. However, for some special vmas, we go ahead and
474 * populate the ptes.
475 *
476 * For vmas that pass the filters, merge/split as appropriate.
477 */
478static int mlock_fixup(struct vma_iterator *vmi, struct vm_area_struct *vma,
479 struct vm_area_struct **prev, unsigned long start,
480 unsigned long end, vm_flags_t newflags)
481{
482 struct mm_struct *mm = vma->vm_mm;
483 int nr_pages;
484 int ret = 0;
485 vm_flags_t oldflags = vma->vm_flags;
486
487 if (newflags == oldflags || (oldflags & VM_SPECIAL) ||
488 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm) ||
489 vma_is_dax(vma) || vma_is_secretmem(vma))
490 /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
491 goto out;
492
493 vma = vma_modify_flags(vmi, *prev, vma, start, end, newflags);
494 if (IS_ERR(vma)) {
495 ret = PTR_ERR(vma);
496 goto out;
497 }
498
499 /*
500 * Keep track of amount of locked VM.
501 */
502 nr_pages = (end - start) >> PAGE_SHIFT;
503 if (!(newflags & VM_LOCKED))
504 nr_pages = -nr_pages;
505 else if (oldflags & VM_LOCKED)
506 nr_pages = 0;
507 mm->locked_vm += nr_pages;
508
509 /*
510 * vm_flags is protected by the mmap_lock held in write mode.
511 * It's okay if try_to_unmap_one unmaps a page just after we
512 * set VM_LOCKED, populate_vma_page_range will bring it back.
513 */
514 if ((newflags & VM_LOCKED) && (oldflags & VM_LOCKED)) {
515 /* No work to do, and mlocking twice would be wrong */
516 vma_start_write(vma);
517 vm_flags_reset(vma, newflags);
518 } else {
519 mlock_vma_pages_range(vma, start, end, newflags);
520 }
521out:
522 *prev = vma;
523 return ret;
524}
525
526static int apply_vma_lock_flags(unsigned long start, size_t len,
527 vm_flags_t flags)
528{
529 unsigned long nstart, end, tmp;
530 struct vm_area_struct *vma, *prev;
531 VMA_ITERATOR(vmi, current->mm, start);
532
533 VM_BUG_ON(offset_in_page(start));
534 VM_BUG_ON(len != PAGE_ALIGN(len));
535 end = start + len;
536 if (end < start)
537 return -EINVAL;
538 if (end == start)
539 return 0;
540 vma = vma_iter_load(&vmi);
541 if (!vma)
542 return -ENOMEM;
543
544 prev = vma_prev(&vmi);
545 if (start > vma->vm_start)
546 prev = vma;
547
548 nstart = start;
549 tmp = vma->vm_start;
550 for_each_vma_range(vmi, vma, end) {
551 int error;
552 vm_flags_t newflags;
553
554 if (vma->vm_start != tmp)
555 return -ENOMEM;
556
557 newflags = vma->vm_flags & ~VM_LOCKED_MASK;
558 newflags |= flags;
559 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
560 tmp = vma->vm_end;
561 if (tmp > end)
562 tmp = end;
563 error = mlock_fixup(&vmi, vma, &prev, nstart, tmp, newflags);
564 if (error)
565 return error;
566 tmp = vma_iter_end(&vmi);
567 nstart = tmp;
568 }
569
570 if (tmp < end)
571 return -ENOMEM;
572
573 return 0;
574}
575
576/*
577 * Go through vma areas and sum size of mlocked
578 * vma pages, as return value.
579 * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT)
580 * is also counted.
581 * Return value: previously mlocked page counts
582 */
583static unsigned long count_mm_mlocked_page_nr(struct mm_struct *mm,
584 unsigned long start, size_t len)
585{
586 struct vm_area_struct *vma;
587 unsigned long count = 0;
588 unsigned long end;
589 VMA_ITERATOR(vmi, mm, start);
590
591 /* Don't overflow past ULONG_MAX */
592 if (unlikely(ULONG_MAX - len < start))
593 end = ULONG_MAX;
594 else
595 end = start + len;
596
597 for_each_vma_range(vmi, vma, end) {
598 if (vma->vm_flags & VM_LOCKED) {
599 if (start > vma->vm_start)
600 count -= (start - vma->vm_start);
601 if (end < vma->vm_end) {
602 count += end - vma->vm_start;
603 break;
604 }
605 count += vma->vm_end - vma->vm_start;
606 }
607 }
608
609 return count >> PAGE_SHIFT;
610}
611
612/*
613 * convert get_user_pages() return value to posix mlock() error
614 */
615static int __mlock_posix_error_return(long retval)
616{
617 if (retval == -EFAULT)
618 retval = -ENOMEM;
619 else if (retval == -ENOMEM)
620 retval = -EAGAIN;
621 return retval;
622}
623
624static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
625{
626 unsigned long locked;
627 unsigned long lock_limit;
628 int error = -ENOMEM;
629
630 start = untagged_addr(start);
631
632 if (!can_do_mlock())
633 return -EPERM;
634
635 len = PAGE_ALIGN(len + (offset_in_page(start)));
636 start &= PAGE_MASK;
637
638 lock_limit = rlimit(RLIMIT_MEMLOCK);
639 lock_limit >>= PAGE_SHIFT;
640 locked = len >> PAGE_SHIFT;
641
642 if (mmap_write_lock_killable(current->mm))
643 return -EINTR;
644
645 locked += current->mm->locked_vm;
646 if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) {
647 /*
648 * It is possible that the regions requested intersect with
649 * previously mlocked areas, that part area in "mm->locked_vm"
650 * should not be counted to new mlock increment count. So check
651 * and adjust locked count if necessary.
652 */
653 locked -= count_mm_mlocked_page_nr(current->mm,
654 start, len);
655 }
656
657 /* check against resource limits */
658 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
659 error = apply_vma_lock_flags(start, len, flags);
660
661 mmap_write_unlock(current->mm);
662 if (error)
663 return error;
664
665 error = __mm_populate(start, len, 0);
666 if (error)
667 return __mlock_posix_error_return(error);
668 return 0;
669}
670
671SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
672{
673 return do_mlock(start, len, VM_LOCKED);
674}
675
676SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
677{
678 vm_flags_t vm_flags = VM_LOCKED;
679
680 if (flags & ~MLOCK_ONFAULT)
681 return -EINVAL;
682
683 if (flags & MLOCK_ONFAULT)
684 vm_flags |= VM_LOCKONFAULT;
685
686 return do_mlock(start, len, vm_flags);
687}
688
689SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
690{
691 int ret;
692
693 start = untagged_addr(start);
694
695 len = PAGE_ALIGN(len + (offset_in_page(start)));
696 start &= PAGE_MASK;
697
698 if (mmap_write_lock_killable(current->mm))
699 return -EINTR;
700 ret = apply_vma_lock_flags(start, len, 0);
701 mmap_write_unlock(current->mm);
702
703 return ret;
704}
705
706/*
707 * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
708 * and translate into the appropriate modifications to mm->def_flags and/or the
709 * flags for all current VMAs.
710 *
711 * There are a couple of subtleties with this. If mlockall() is called multiple
712 * times with different flags, the values do not necessarily stack. If mlockall
713 * is called once including the MCL_FUTURE flag and then a second time without
714 * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
715 */
716static int apply_mlockall_flags(int flags)
717{
718 VMA_ITERATOR(vmi, current->mm, 0);
719 struct vm_area_struct *vma, *prev = NULL;
720 vm_flags_t to_add = 0;
721
722 current->mm->def_flags &= ~VM_LOCKED_MASK;
723 if (flags & MCL_FUTURE) {
724 current->mm->def_flags |= VM_LOCKED;
725
726 if (flags & MCL_ONFAULT)
727 current->mm->def_flags |= VM_LOCKONFAULT;
728
729 if (!(flags & MCL_CURRENT))
730 goto out;
731 }
732
733 if (flags & MCL_CURRENT) {
734 to_add |= VM_LOCKED;
735 if (flags & MCL_ONFAULT)
736 to_add |= VM_LOCKONFAULT;
737 }
738
739 for_each_vma(vmi, vma) {
740 vm_flags_t newflags;
741
742 newflags = vma->vm_flags & ~VM_LOCKED_MASK;
743 newflags |= to_add;
744
745 /* Ignore errors */
746 mlock_fixup(&vmi, vma, &prev, vma->vm_start, vma->vm_end,
747 newflags);
748 cond_resched();
749 }
750out:
751 return 0;
752}
753
754SYSCALL_DEFINE1(mlockall, int, flags)
755{
756 unsigned long lock_limit;
757 int ret;
758
759 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)) ||
760 flags == MCL_ONFAULT)
761 return -EINVAL;
762
763 if (!can_do_mlock())
764 return -EPERM;
765
766 lock_limit = rlimit(RLIMIT_MEMLOCK);
767 lock_limit >>= PAGE_SHIFT;
768
769 if (mmap_write_lock_killable(current->mm))
770 return -EINTR;
771
772 ret = -ENOMEM;
773 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
774 capable(CAP_IPC_LOCK))
775 ret = apply_mlockall_flags(flags);
776 mmap_write_unlock(current->mm);
777 if (!ret && (flags & MCL_CURRENT))
778 mm_populate(0, TASK_SIZE);
779
780 return ret;
781}
782
783SYSCALL_DEFINE0(munlockall)
784{
785 int ret;
786
787 if (mmap_write_lock_killable(current->mm))
788 return -EINTR;
789 ret = apply_mlockall_flags(0);
790 mmap_write_unlock(current->mm);
791 return ret;
792}
793
794/*
795 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
796 * shm segments) get accounted against the user_struct instead.
797 */
798static DEFINE_SPINLOCK(shmlock_user_lock);
799
800int user_shm_lock(size_t size, struct ucounts *ucounts)
801{
802 unsigned long lock_limit, locked;
803 long memlock;
804 int allowed = 0;
805
806 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
807 lock_limit = rlimit(RLIMIT_MEMLOCK);
808 if (lock_limit != RLIM_INFINITY)
809 lock_limit >>= PAGE_SHIFT;
810 spin_lock(&shmlock_user_lock);
811 memlock = inc_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
812
813 if ((memlock == LONG_MAX || memlock > lock_limit) && !capable(CAP_IPC_LOCK)) {
814 dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
815 goto out;
816 }
817 if (!get_ucounts(ucounts)) {
818 dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
819 allowed = 0;
820 goto out;
821 }
822 allowed = 1;
823out:
824 spin_unlock(&shmlock_user_lock);
825 return allowed;
826}
827
828void user_shm_unlock(size_t size, struct ucounts *ucounts)
829{
830 spin_lock(&shmlock_user_lock);
831 dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, (size + PAGE_SIZE - 1) >> PAGE_SHIFT);
832 spin_unlock(&shmlock_user_lock);
833 put_ucounts(ucounts);
834}
1/*
2 * linux/mm/mlock.c
3 *
4 * (C) Copyright 1995 Linus Torvalds
5 * (C) Copyright 2002 Christoph Hellwig
6 */
7
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/pagevec.h>
15#include <linux/mempolicy.h>
16#include <linux/syscalls.h>
17#include <linux/sched.h>
18#include <linux/export.h>
19#include <linux/rmap.h>
20#include <linux/mmzone.h>
21#include <linux/hugetlb.h>
22#include <linux/memcontrol.h>
23#include <linux/mm_inline.h>
24
25#include "internal.h"
26
27bool can_do_mlock(void)
28{
29 if (rlimit(RLIMIT_MEMLOCK) != 0)
30 return true;
31 if (capable(CAP_IPC_LOCK))
32 return true;
33 return false;
34}
35EXPORT_SYMBOL(can_do_mlock);
36
37/*
38 * Mlocked pages are marked with PageMlocked() flag for efficient testing
39 * in vmscan and, possibly, the fault path; and to support semi-accurate
40 * statistics.
41 *
42 * An mlocked page [PageMlocked(page)] is unevictable. As such, it will
43 * be placed on the LRU "unevictable" list, rather than the [in]active lists.
44 * The unevictable list is an LRU sibling list to the [in]active lists.
45 * PageUnevictable is set to indicate the unevictable state.
46 *
47 * When lazy mlocking via vmscan, it is important to ensure that the
48 * vma's VM_LOCKED status is not concurrently being modified, otherwise we
49 * may have mlocked a page that is being munlocked. So lazy mlock must take
50 * the mmap_sem for read, and verify that the vma really is locked
51 * (see mm/rmap.c).
52 */
53
54/*
55 * LRU accounting for clear_page_mlock()
56 */
57void clear_page_mlock(struct page *page)
58{
59 if (!TestClearPageMlocked(page))
60 return;
61
62 mod_zone_page_state(page_zone(page), NR_MLOCK,
63 -hpage_nr_pages(page));
64 count_vm_event(UNEVICTABLE_PGCLEARED);
65 if (!isolate_lru_page(page)) {
66 putback_lru_page(page);
67 } else {
68 /*
69 * We lost the race. the page already moved to evictable list.
70 */
71 if (PageUnevictable(page))
72 count_vm_event(UNEVICTABLE_PGSTRANDED);
73 }
74}
75
76/*
77 * Mark page as mlocked if not already.
78 * If page on LRU, isolate and putback to move to unevictable list.
79 */
80void mlock_vma_page(struct page *page)
81{
82 /* Serialize with page migration */
83 BUG_ON(!PageLocked(page));
84
85 VM_BUG_ON_PAGE(PageTail(page), page);
86 VM_BUG_ON_PAGE(PageCompound(page) && PageDoubleMap(page), page);
87
88 if (!TestSetPageMlocked(page)) {
89 mod_zone_page_state(page_zone(page), NR_MLOCK,
90 hpage_nr_pages(page));
91 count_vm_event(UNEVICTABLE_PGMLOCKED);
92 if (!isolate_lru_page(page))
93 putback_lru_page(page);
94 }
95}
96
97/*
98 * Isolate a page from LRU with optional get_page() pin.
99 * Assumes lru_lock already held and page already pinned.
100 */
101static bool __munlock_isolate_lru_page(struct page *page, bool getpage)
102{
103 if (PageLRU(page)) {
104 struct lruvec *lruvec;
105
106 lruvec = mem_cgroup_page_lruvec(page, page_zone(page));
107 if (getpage)
108 get_page(page);
109 ClearPageLRU(page);
110 del_page_from_lru_list(page, lruvec, page_lru(page));
111 return true;
112 }
113
114 return false;
115}
116
117/*
118 * Finish munlock after successful page isolation
119 *
120 * Page must be locked. This is a wrapper for try_to_munlock()
121 * and putback_lru_page() with munlock accounting.
122 */
123static void __munlock_isolated_page(struct page *page)
124{
125 int ret = SWAP_AGAIN;
126
127 /*
128 * Optimization: if the page was mapped just once, that's our mapping
129 * and we don't need to check all the other vmas.
130 */
131 if (page_mapcount(page) > 1)
132 ret = try_to_munlock(page);
133
134 /* Did try_to_unlock() succeed or punt? */
135 if (ret != SWAP_MLOCK)
136 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
137
138 putback_lru_page(page);
139}
140
141/*
142 * Accounting for page isolation fail during munlock
143 *
144 * Performs accounting when page isolation fails in munlock. There is nothing
145 * else to do because it means some other task has already removed the page
146 * from the LRU. putback_lru_page() will take care of removing the page from
147 * the unevictable list, if necessary. vmscan [page_referenced()] will move
148 * the page back to the unevictable list if some other vma has it mlocked.
149 */
150static void __munlock_isolation_failed(struct page *page)
151{
152 if (PageUnevictable(page))
153 __count_vm_event(UNEVICTABLE_PGSTRANDED);
154 else
155 __count_vm_event(UNEVICTABLE_PGMUNLOCKED);
156}
157
158/**
159 * munlock_vma_page - munlock a vma page
160 * @page - page to be unlocked, either a normal page or THP page head
161 *
162 * returns the size of the page as a page mask (0 for normal page,
163 * HPAGE_PMD_NR - 1 for THP head page)
164 *
165 * called from munlock()/munmap() path with page supposedly on the LRU.
166 * When we munlock a page, because the vma where we found the page is being
167 * munlock()ed or munmap()ed, we want to check whether other vmas hold the
168 * page locked so that we can leave it on the unevictable lru list and not
169 * bother vmscan with it. However, to walk the page's rmap list in
170 * try_to_munlock() we must isolate the page from the LRU. If some other
171 * task has removed the page from the LRU, we won't be able to do that.
172 * So we clear the PageMlocked as we might not get another chance. If we
173 * can't isolate the page, we leave it for putback_lru_page() and vmscan
174 * [page_referenced()/try_to_unmap()] to deal with.
175 */
176unsigned int munlock_vma_page(struct page *page)
177{
178 int nr_pages;
179 struct zone *zone = page_zone(page);
180
181 /* For try_to_munlock() and to serialize with page migration */
182 BUG_ON(!PageLocked(page));
183
184 VM_BUG_ON_PAGE(PageTail(page), page);
185
186 /*
187 * Serialize with any parallel __split_huge_page_refcount() which
188 * might otherwise copy PageMlocked to part of the tail pages before
189 * we clear it in the head page. It also stabilizes hpage_nr_pages().
190 */
191 spin_lock_irq(&zone->lru_lock);
192
193 nr_pages = hpage_nr_pages(page);
194 if (!TestClearPageMlocked(page))
195 goto unlock_out;
196
197 __mod_zone_page_state(zone, NR_MLOCK, -nr_pages);
198
199 if (__munlock_isolate_lru_page(page, true)) {
200 spin_unlock_irq(&zone->lru_lock);
201 __munlock_isolated_page(page);
202 goto out;
203 }
204 __munlock_isolation_failed(page);
205
206unlock_out:
207 spin_unlock_irq(&zone->lru_lock);
208
209out:
210 return nr_pages - 1;
211}
212
213/*
214 * convert get_user_pages() return value to posix mlock() error
215 */
216static int __mlock_posix_error_return(long retval)
217{
218 if (retval == -EFAULT)
219 retval = -ENOMEM;
220 else if (retval == -ENOMEM)
221 retval = -EAGAIN;
222 return retval;
223}
224
225/*
226 * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec()
227 *
228 * The fast path is available only for evictable pages with single mapping.
229 * Then we can bypass the per-cpu pvec and get better performance.
230 * when mapcount > 1 we need try_to_munlock() which can fail.
231 * when !page_evictable(), we need the full redo logic of putback_lru_page to
232 * avoid leaving evictable page in unevictable list.
233 *
234 * In case of success, @page is added to @pvec and @pgrescued is incremented
235 * in case that the page was previously unevictable. @page is also unlocked.
236 */
237static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec,
238 int *pgrescued)
239{
240 VM_BUG_ON_PAGE(PageLRU(page), page);
241 VM_BUG_ON_PAGE(!PageLocked(page), page);
242
243 if (page_mapcount(page) <= 1 && page_evictable(page)) {
244 pagevec_add(pvec, page);
245 if (TestClearPageUnevictable(page))
246 (*pgrescued)++;
247 unlock_page(page);
248 return true;
249 }
250
251 return false;
252}
253
254/*
255 * Putback multiple evictable pages to the LRU
256 *
257 * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of
258 * the pages might have meanwhile become unevictable but that is OK.
259 */
260static void __putback_lru_fast(struct pagevec *pvec, int pgrescued)
261{
262 count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec));
263 /*
264 *__pagevec_lru_add() calls release_pages() so we don't call
265 * put_page() explicitly
266 */
267 __pagevec_lru_add(pvec);
268 count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued);
269}
270
271/*
272 * Munlock a batch of pages from the same zone
273 *
274 * The work is split to two main phases. First phase clears the Mlocked flag
275 * and attempts to isolate the pages, all under a single zone lru lock.
276 * The second phase finishes the munlock only for pages where isolation
277 * succeeded.
278 *
279 * Note that the pagevec may be modified during the process.
280 */
281static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone)
282{
283 int i;
284 int nr = pagevec_count(pvec);
285 int delta_munlocked;
286 struct pagevec pvec_putback;
287 int pgrescued = 0;
288
289 pagevec_init(&pvec_putback, 0);
290
291 /* Phase 1: page isolation */
292 spin_lock_irq(&zone->lru_lock);
293 for (i = 0; i < nr; i++) {
294 struct page *page = pvec->pages[i];
295
296 if (TestClearPageMlocked(page)) {
297 /*
298 * We already have pin from follow_page_mask()
299 * so we can spare the get_page() here.
300 */
301 if (__munlock_isolate_lru_page(page, false))
302 continue;
303 else
304 __munlock_isolation_failed(page);
305 }
306
307 /*
308 * We won't be munlocking this page in the next phase
309 * but we still need to release the follow_page_mask()
310 * pin. We cannot do it under lru_lock however. If it's
311 * the last pin, __page_cache_release() would deadlock.
312 */
313 pagevec_add(&pvec_putback, pvec->pages[i]);
314 pvec->pages[i] = NULL;
315 }
316 delta_munlocked = -nr + pagevec_count(&pvec_putback);
317 __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked);
318 spin_unlock_irq(&zone->lru_lock);
319
320 /* Now we can release pins of pages that we are not munlocking */
321 pagevec_release(&pvec_putback);
322
323 /* Phase 2: page munlock */
324 for (i = 0; i < nr; i++) {
325 struct page *page = pvec->pages[i];
326
327 if (page) {
328 lock_page(page);
329 if (!__putback_lru_fast_prepare(page, &pvec_putback,
330 &pgrescued)) {
331 /*
332 * Slow path. We don't want to lose the last
333 * pin before unlock_page()
334 */
335 get_page(page); /* for putback_lru_page() */
336 __munlock_isolated_page(page);
337 unlock_page(page);
338 put_page(page); /* from follow_page_mask() */
339 }
340 }
341 }
342
343 /*
344 * Phase 3: page putback for pages that qualified for the fast path
345 * This will also call put_page() to return pin from follow_page_mask()
346 */
347 if (pagevec_count(&pvec_putback))
348 __putback_lru_fast(&pvec_putback, pgrescued);
349}
350
351/*
352 * Fill up pagevec for __munlock_pagevec using pte walk
353 *
354 * The function expects that the struct page corresponding to @start address is
355 * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone.
356 *
357 * The rest of @pvec is filled by subsequent pages within the same pmd and same
358 * zone, as long as the pte's are present and vm_normal_page() succeeds. These
359 * pages also get pinned.
360 *
361 * Returns the address of the next page that should be scanned. This equals
362 * @start + PAGE_SIZE when no page could be added by the pte walk.
363 */
364static unsigned long __munlock_pagevec_fill(struct pagevec *pvec,
365 struct vm_area_struct *vma, int zoneid, unsigned long start,
366 unsigned long end)
367{
368 pte_t *pte;
369 spinlock_t *ptl;
370
371 /*
372 * Initialize pte walk starting at the already pinned page where we
373 * are sure that there is a pte, as it was pinned under the same
374 * mmap_sem write op.
375 */
376 pte = get_locked_pte(vma->vm_mm, start, &ptl);
377 /* Make sure we do not cross the page table boundary */
378 end = pgd_addr_end(start, end);
379 end = pud_addr_end(start, end);
380 end = pmd_addr_end(start, end);
381
382 /* The page next to the pinned page is the first we will try to get */
383 start += PAGE_SIZE;
384 while (start < end) {
385 struct page *page = NULL;
386 pte++;
387 if (pte_present(*pte))
388 page = vm_normal_page(vma, start, *pte);
389 /*
390 * Break if page could not be obtained or the page's node+zone does not
391 * match
392 */
393 if (!page || page_zone_id(page) != zoneid)
394 break;
395
396 /*
397 * Do not use pagevec for PTE-mapped THP,
398 * munlock_vma_pages_range() will handle them.
399 */
400 if (PageTransCompound(page))
401 break;
402
403 get_page(page);
404 /*
405 * Increase the address that will be returned *before* the
406 * eventual break due to pvec becoming full by adding the page
407 */
408 start += PAGE_SIZE;
409 if (pagevec_add(pvec, page) == 0)
410 break;
411 }
412 pte_unmap_unlock(pte, ptl);
413 return start;
414}
415
416/*
417 * munlock_vma_pages_range() - munlock all pages in the vma range.'
418 * @vma - vma containing range to be munlock()ed.
419 * @start - start address in @vma of the range
420 * @end - end of range in @vma.
421 *
422 * For mremap(), munmap() and exit().
423 *
424 * Called with @vma VM_LOCKED.
425 *
426 * Returns with VM_LOCKED cleared. Callers must be prepared to
427 * deal with this.
428 *
429 * We don't save and restore VM_LOCKED here because pages are
430 * still on lru. In unmap path, pages might be scanned by reclaim
431 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
432 * free them. This will result in freeing mlocked pages.
433 */
434void munlock_vma_pages_range(struct vm_area_struct *vma,
435 unsigned long start, unsigned long end)
436{
437 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
438
439 while (start < end) {
440 struct page *page;
441 unsigned int page_mask;
442 unsigned long page_increm;
443 struct pagevec pvec;
444 struct zone *zone;
445 int zoneid;
446
447 pagevec_init(&pvec, 0);
448 /*
449 * Although FOLL_DUMP is intended for get_dump_page(),
450 * it just so happens that its special treatment of the
451 * ZERO_PAGE (returning an error instead of doing get_page)
452 * suits munlock very well (and if somehow an abnormal page
453 * has sneaked into the range, we won't oops here: great).
454 */
455 page = follow_page_mask(vma, start, FOLL_GET | FOLL_DUMP,
456 &page_mask);
457
458 if (page && !IS_ERR(page)) {
459 if (PageTransTail(page)) {
460 VM_BUG_ON_PAGE(PageMlocked(page), page);
461 put_page(page); /* follow_page_mask() */
462 } else if (PageTransHuge(page)) {
463 lock_page(page);
464 /*
465 * Any THP page found by follow_page_mask() may
466 * have gotten split before reaching
467 * munlock_vma_page(), so we need to recompute
468 * the page_mask here.
469 */
470 page_mask = munlock_vma_page(page);
471 unlock_page(page);
472 put_page(page); /* follow_page_mask() */
473 } else {
474 /*
475 * Non-huge pages are handled in batches via
476 * pagevec. The pin from follow_page_mask()
477 * prevents them from collapsing by THP.
478 */
479 pagevec_add(&pvec, page);
480 zone = page_zone(page);
481 zoneid = page_zone_id(page);
482
483 /*
484 * Try to fill the rest of pagevec using fast
485 * pte walk. This will also update start to
486 * the next page to process. Then munlock the
487 * pagevec.
488 */
489 start = __munlock_pagevec_fill(&pvec, vma,
490 zoneid, start, end);
491 __munlock_pagevec(&pvec, zone);
492 goto next;
493 }
494 }
495 page_increm = 1 + page_mask;
496 start += page_increm * PAGE_SIZE;
497next:
498 cond_resched();
499 }
500}
501
502/*
503 * mlock_fixup - handle mlock[all]/munlock[all] requests.
504 *
505 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
506 * munlock is a no-op. However, for some special vmas, we go ahead and
507 * populate the ptes.
508 *
509 * For vmas that pass the filters, merge/split as appropriate.
510 */
511static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
512 unsigned long start, unsigned long end, vm_flags_t newflags)
513{
514 struct mm_struct *mm = vma->vm_mm;
515 pgoff_t pgoff;
516 int nr_pages;
517 int ret = 0;
518 int lock = !!(newflags & VM_LOCKED);
519
520 if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
521 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
522 /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
523 goto out;
524
525 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
526 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
527 vma->vm_file, pgoff, vma_policy(vma),
528 vma->vm_userfaultfd_ctx);
529 if (*prev) {
530 vma = *prev;
531 goto success;
532 }
533
534 if (start != vma->vm_start) {
535 ret = split_vma(mm, vma, start, 1);
536 if (ret)
537 goto out;
538 }
539
540 if (end != vma->vm_end) {
541 ret = split_vma(mm, vma, end, 0);
542 if (ret)
543 goto out;
544 }
545
546success:
547 /*
548 * Keep track of amount of locked VM.
549 */
550 nr_pages = (end - start) >> PAGE_SHIFT;
551 if (!lock)
552 nr_pages = -nr_pages;
553 mm->locked_vm += nr_pages;
554
555 /*
556 * vm_flags is protected by the mmap_sem held in write mode.
557 * It's okay if try_to_unmap_one unmaps a page just after we
558 * set VM_LOCKED, populate_vma_page_range will bring it back.
559 */
560
561 if (lock)
562 vma->vm_flags = newflags;
563 else
564 munlock_vma_pages_range(vma, start, end);
565
566out:
567 *prev = vma;
568 return ret;
569}
570
571static int apply_vma_lock_flags(unsigned long start, size_t len,
572 vm_flags_t flags)
573{
574 unsigned long nstart, end, tmp;
575 struct vm_area_struct * vma, * prev;
576 int error;
577
578 VM_BUG_ON(offset_in_page(start));
579 VM_BUG_ON(len != PAGE_ALIGN(len));
580 end = start + len;
581 if (end < start)
582 return -EINVAL;
583 if (end == start)
584 return 0;
585 vma = find_vma(current->mm, start);
586 if (!vma || vma->vm_start > start)
587 return -ENOMEM;
588
589 prev = vma->vm_prev;
590 if (start > vma->vm_start)
591 prev = vma;
592
593 for (nstart = start ; ; ) {
594 vm_flags_t newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
595
596 newflags |= flags;
597
598 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
599 tmp = vma->vm_end;
600 if (tmp > end)
601 tmp = end;
602 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
603 if (error)
604 break;
605 nstart = tmp;
606 if (nstart < prev->vm_end)
607 nstart = prev->vm_end;
608 if (nstart >= end)
609 break;
610
611 vma = prev->vm_next;
612 if (!vma || vma->vm_start != nstart) {
613 error = -ENOMEM;
614 break;
615 }
616 }
617 return error;
618}
619
620static int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
621{
622 unsigned long locked;
623 unsigned long lock_limit;
624 int error = -ENOMEM;
625
626 if (!can_do_mlock())
627 return -EPERM;
628
629 lru_add_drain_all(); /* flush pagevec */
630
631 len = PAGE_ALIGN(len + (offset_in_page(start)));
632 start &= PAGE_MASK;
633
634 lock_limit = rlimit(RLIMIT_MEMLOCK);
635 lock_limit >>= PAGE_SHIFT;
636 locked = len >> PAGE_SHIFT;
637
638 down_write(¤t->mm->mmap_sem);
639
640 locked += current->mm->locked_vm;
641
642 /* check against resource limits */
643 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
644 error = apply_vma_lock_flags(start, len, flags);
645
646 up_write(¤t->mm->mmap_sem);
647 if (error)
648 return error;
649
650 error = __mm_populate(start, len, 0);
651 if (error)
652 return __mlock_posix_error_return(error);
653 return 0;
654}
655
656SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
657{
658 return do_mlock(start, len, VM_LOCKED);
659}
660
661SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
662{
663 vm_flags_t vm_flags = VM_LOCKED;
664
665 if (flags & ~MLOCK_ONFAULT)
666 return -EINVAL;
667
668 if (flags & MLOCK_ONFAULT)
669 vm_flags |= VM_LOCKONFAULT;
670
671 return do_mlock(start, len, vm_flags);
672}
673
674SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
675{
676 int ret;
677
678 len = PAGE_ALIGN(len + (offset_in_page(start)));
679 start &= PAGE_MASK;
680
681 down_write(¤t->mm->mmap_sem);
682 ret = apply_vma_lock_flags(start, len, 0);
683 up_write(¤t->mm->mmap_sem);
684
685 return ret;
686}
687
688/*
689 * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
690 * and translate into the appropriate modifications to mm->def_flags and/or the
691 * flags for all current VMAs.
692 *
693 * There are a couple of subtleties with this. If mlockall() is called multiple
694 * times with different flags, the values do not necessarily stack. If mlockall
695 * is called once including the MCL_FUTURE flag and then a second time without
696 * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
697 */
698static int apply_mlockall_flags(int flags)
699{
700 struct vm_area_struct * vma, * prev = NULL;
701 vm_flags_t to_add = 0;
702
703 current->mm->def_flags &= VM_LOCKED_CLEAR_MASK;
704 if (flags & MCL_FUTURE) {
705 current->mm->def_flags |= VM_LOCKED;
706
707 if (flags & MCL_ONFAULT)
708 current->mm->def_flags |= VM_LOCKONFAULT;
709
710 if (!(flags & MCL_CURRENT))
711 goto out;
712 }
713
714 if (flags & MCL_CURRENT) {
715 to_add |= VM_LOCKED;
716 if (flags & MCL_ONFAULT)
717 to_add |= VM_LOCKONFAULT;
718 }
719
720 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
721 vm_flags_t newflags;
722
723 newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
724 newflags |= to_add;
725
726 /* Ignore errors */
727 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
728 cond_resched_rcu_qs();
729 }
730out:
731 return 0;
732}
733
734SYSCALL_DEFINE1(mlockall, int, flags)
735{
736 unsigned long lock_limit;
737 int ret;
738
739 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)))
740 return -EINVAL;
741
742 if (!can_do_mlock())
743 return -EPERM;
744
745 if (flags & MCL_CURRENT)
746 lru_add_drain_all(); /* flush pagevec */
747
748 lock_limit = rlimit(RLIMIT_MEMLOCK);
749 lock_limit >>= PAGE_SHIFT;
750
751 ret = -ENOMEM;
752 down_write(¤t->mm->mmap_sem);
753
754 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
755 capable(CAP_IPC_LOCK))
756 ret = apply_mlockall_flags(flags);
757 up_write(¤t->mm->mmap_sem);
758 if (!ret && (flags & MCL_CURRENT))
759 mm_populate(0, TASK_SIZE);
760
761 return ret;
762}
763
764SYSCALL_DEFINE0(munlockall)
765{
766 int ret;
767
768 down_write(¤t->mm->mmap_sem);
769 ret = apply_mlockall_flags(0);
770 up_write(¤t->mm->mmap_sem);
771 return ret;
772}
773
774/*
775 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
776 * shm segments) get accounted against the user_struct instead.
777 */
778static DEFINE_SPINLOCK(shmlock_user_lock);
779
780int user_shm_lock(size_t size, struct user_struct *user)
781{
782 unsigned long lock_limit, locked;
783 int allowed = 0;
784
785 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
786 lock_limit = rlimit(RLIMIT_MEMLOCK);
787 if (lock_limit == RLIM_INFINITY)
788 allowed = 1;
789 lock_limit >>= PAGE_SHIFT;
790 spin_lock(&shmlock_user_lock);
791 if (!allowed &&
792 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
793 goto out;
794 get_uid(user);
795 user->locked_shm += locked;
796 allowed = 1;
797out:
798 spin_unlock(&shmlock_user_lock);
799 return allowed;
800}
801
802void user_shm_unlock(size_t size, struct user_struct *user)
803{
804 spin_lock(&shmlock_user_lock);
805 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
806 spin_unlock(&shmlock_user_lock);
807 free_uid(user);
808}