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);
210}
211
212void mlock_drain_local(void)
213{
214 struct folio_batch *fbatch;
215
216 local_lock(&mlock_fbatch.lock);
217 fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
218 if (folio_batch_count(fbatch))
219 mlock_folio_batch(fbatch);
220 local_unlock(&mlock_fbatch.lock);
221}
222
223void mlock_drain_remote(int cpu)
224{
225 struct folio_batch *fbatch;
226
227 WARN_ON_ONCE(cpu_online(cpu));
228 fbatch = &per_cpu(mlock_fbatch.fbatch, cpu);
229 if (folio_batch_count(fbatch))
230 mlock_folio_batch(fbatch);
231}
232
233bool need_mlock_drain(int cpu)
234{
235 return folio_batch_count(&per_cpu(mlock_fbatch.fbatch, cpu));
236}
237
238/**
239 * mlock_folio - mlock a folio already on (or temporarily off) LRU
240 * @folio: folio to be mlocked.
241 */
242void mlock_folio(struct folio *folio)
243{
244 struct folio_batch *fbatch;
245
246 local_lock(&mlock_fbatch.lock);
247 fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
248
249 if (!folio_test_set_mlocked(folio)) {
250 int nr_pages = folio_nr_pages(folio);
251
252 zone_stat_mod_folio(folio, NR_MLOCK, nr_pages);
253 __count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
254 }
255
256 folio_get(folio);
257 if (!folio_batch_add(fbatch, mlock_lru(folio)) ||
258 folio_test_large(folio) || lru_cache_disabled())
259 mlock_folio_batch(fbatch);
260 local_unlock(&mlock_fbatch.lock);
261}
262
263/**
264 * mlock_new_folio - mlock a newly allocated folio not yet on LRU
265 * @folio: folio to be mlocked, either normal or a THP head.
266 */
267void mlock_new_folio(struct folio *folio)
268{
269 struct folio_batch *fbatch;
270 int nr_pages = folio_nr_pages(folio);
271
272 local_lock(&mlock_fbatch.lock);
273 fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
274 folio_set_mlocked(folio);
275
276 zone_stat_mod_folio(folio, NR_MLOCK, nr_pages);
277 __count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
278
279 folio_get(folio);
280 if (!folio_batch_add(fbatch, mlock_new(folio)) ||
281 folio_test_large(folio) || lru_cache_disabled())
282 mlock_folio_batch(fbatch);
283 local_unlock(&mlock_fbatch.lock);
284}
285
286/**
287 * munlock_folio - munlock a folio
288 * @folio: folio to be munlocked, either normal or a THP head.
289 */
290void munlock_folio(struct folio *folio)
291{
292 struct folio_batch *fbatch;
293
294 local_lock(&mlock_fbatch.lock);
295 fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
296 /*
297 * folio_test_clear_mlocked(folio) must be left to __munlock_folio(),
298 * which will check whether the folio is multiply mlocked.
299 */
300 folio_get(folio);
301 if (!folio_batch_add(fbatch, folio) ||
302 folio_test_large(folio) || lru_cache_disabled())
303 mlock_folio_batch(fbatch);
304 local_unlock(&mlock_fbatch.lock);
305}
306
307static inline unsigned int folio_mlock_step(struct folio *folio,
308 pte_t *pte, unsigned long addr, unsigned long end)
309{
310 const fpb_t fpb_flags = FPB_IGNORE_DIRTY | FPB_IGNORE_SOFT_DIRTY;
311 unsigned int count = (end - addr) >> PAGE_SHIFT;
312 pte_t ptent = ptep_get(pte);
313
314 if (!folio_test_large(folio))
315 return 1;
316
317 return folio_pte_batch(folio, addr, pte, ptent, count, fpb_flags, NULL,
318 NULL, NULL);
319}
320
321static inline bool allow_mlock_munlock(struct folio *folio,
322 struct vm_area_struct *vma, unsigned long start,
323 unsigned long end, unsigned int step)
324{
325 /*
326 * For unlock, allow munlock large folio which is partially
327 * mapped to VMA. As it's possible that large folio is
328 * mlocked and VMA is split later.
329 *
330 * During memory pressure, such kind of large folio can
331 * be split. And the pages are not in VM_LOCKed VMA
332 * can be reclaimed.
333 */
334 if (!(vma->vm_flags & VM_LOCKED))
335 return true;
336
337 /* folio_within_range() cannot take KSM, but any small folio is OK */
338 if (!folio_test_large(folio))
339 return true;
340
341 /* folio not in range [start, end), skip mlock */
342 if (!folio_within_range(folio, vma, start, end))
343 return false;
344
345 /* folio is not fully mapped, skip mlock */
346 if (step != folio_nr_pages(folio))
347 return false;
348
349 return true;
350}
351
352static int mlock_pte_range(pmd_t *pmd, unsigned long addr,
353 unsigned long end, struct mm_walk *walk)
354
355{
356 struct vm_area_struct *vma = walk->vma;
357 spinlock_t *ptl;
358 pte_t *start_pte, *pte;
359 pte_t ptent;
360 struct folio *folio;
361 unsigned int step = 1;
362 unsigned long start = addr;
363
364 ptl = pmd_trans_huge_lock(pmd, vma);
365 if (ptl) {
366 if (!pmd_present(*pmd))
367 goto out;
368 if (is_huge_zero_pmd(*pmd))
369 goto out;
370 folio = pmd_folio(*pmd);
371 if (vma->vm_flags & VM_LOCKED)
372 mlock_folio(folio);
373 else
374 munlock_folio(folio);
375 goto out;
376 }
377
378 start_pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
379 if (!start_pte) {
380 walk->action = ACTION_AGAIN;
381 return 0;
382 }
383
384 for (pte = start_pte; addr != end; pte++, addr += PAGE_SIZE) {
385 ptent = ptep_get(pte);
386 if (!pte_present(ptent))
387 continue;
388 folio = vm_normal_folio(vma, addr, ptent);
389 if (!folio || folio_is_zone_device(folio))
390 continue;
391
392 step = folio_mlock_step(folio, pte, addr, end);
393 if (!allow_mlock_munlock(folio, vma, start, end, step))
394 goto next_entry;
395
396 if (vma->vm_flags & VM_LOCKED)
397 mlock_folio(folio);
398 else
399 munlock_folio(folio);
400
401next_entry:
402 pte += step - 1;
403 addr += (step - 1) << PAGE_SHIFT;
404 }
405 pte_unmap(start_pte);
406out:
407 spin_unlock(ptl);
408 cond_resched();
409 return 0;
410}
411
412/*
413 * mlock_vma_pages_range() - mlock any pages already in the range,
414 * or munlock all pages in the range.
415 * @vma - vma containing range to be mlock()ed or munlock()ed
416 * @start - start address in @vma of the range
417 * @end - end of range in @vma
418 * @newflags - the new set of flags for @vma.
419 *
420 * Called for mlock(), mlock2() and mlockall(), to set @vma VM_LOCKED;
421 * called for munlock() and munlockall(), to clear VM_LOCKED from @vma.
422 */
423static void mlock_vma_pages_range(struct vm_area_struct *vma,
424 unsigned long start, unsigned long end, vm_flags_t newflags)
425{
426 static const struct mm_walk_ops mlock_walk_ops = {
427 .pmd_entry = mlock_pte_range,
428 .walk_lock = PGWALK_WRLOCK_VERIFY,
429 };
430
431 /*
432 * There is a slight chance that concurrent page migration,
433 * or page reclaim finding a page of this now-VM_LOCKED vma,
434 * will call mlock_vma_folio() and raise page's mlock_count:
435 * double counting, leaving the page unevictable indefinitely.
436 * Communicate this danger to mlock_vma_folio() with VM_IO,
437 * which is a VM_SPECIAL flag not allowed on VM_LOCKED vmas.
438 * mmap_lock is held in write mode here, so this weird
439 * combination should not be visible to other mmap_lock users;
440 * but WRITE_ONCE so rmap walkers must see VM_IO if VM_LOCKED.
441 */
442 if (newflags & VM_LOCKED)
443 newflags |= VM_IO;
444 vma_start_write(vma);
445 vm_flags_reset_once(vma, newflags);
446
447 lru_add_drain();
448 walk_page_range(vma->vm_mm, start, end, &mlock_walk_ops, NULL);
449 lru_add_drain();
450
451 if (newflags & VM_IO) {
452 newflags &= ~VM_IO;
453 vm_flags_reset_once(vma, newflags);
454 }
455}
456
457/*
458 * mlock_fixup - handle mlock[all]/munlock[all] requests.
459 *
460 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
461 * munlock is a no-op. However, for some special vmas, we go ahead and
462 * populate the ptes.
463 *
464 * For vmas that pass the filters, merge/split as appropriate.
465 */
466static int mlock_fixup(struct vma_iterator *vmi, struct vm_area_struct *vma,
467 struct vm_area_struct **prev, unsigned long start,
468 unsigned long end, vm_flags_t newflags)
469{
470 struct mm_struct *mm = vma->vm_mm;
471 int nr_pages;
472 int ret = 0;
473 vm_flags_t oldflags = vma->vm_flags;
474
475 if (newflags == oldflags || (oldflags & VM_SPECIAL) ||
476 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm) ||
477 vma_is_dax(vma) || vma_is_secretmem(vma) || (oldflags & VM_DROPPABLE))
478 /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
479 goto out;
480
481 vma = vma_modify_flags(vmi, *prev, vma, start, end, newflags);
482 if (IS_ERR(vma)) {
483 ret = PTR_ERR(vma);
484 goto out;
485 }
486
487 /*
488 * Keep track of amount of locked VM.
489 */
490 nr_pages = (end - start) >> PAGE_SHIFT;
491 if (!(newflags & VM_LOCKED))
492 nr_pages = -nr_pages;
493 else if (oldflags & VM_LOCKED)
494 nr_pages = 0;
495 mm->locked_vm += nr_pages;
496
497 /*
498 * vm_flags is protected by the mmap_lock held in write mode.
499 * It's okay if try_to_unmap_one unmaps a page just after we
500 * set VM_LOCKED, populate_vma_page_range will bring it back.
501 */
502 if ((newflags & VM_LOCKED) && (oldflags & VM_LOCKED)) {
503 /* No work to do, and mlocking twice would be wrong */
504 vma_start_write(vma);
505 vm_flags_reset(vma, newflags);
506 } else {
507 mlock_vma_pages_range(vma, start, end, newflags);
508 }
509out:
510 *prev = vma;
511 return ret;
512}
513
514static int apply_vma_lock_flags(unsigned long start, size_t len,
515 vm_flags_t flags)
516{
517 unsigned long nstart, end, tmp;
518 struct vm_area_struct *vma, *prev;
519 VMA_ITERATOR(vmi, current->mm, start);
520
521 VM_BUG_ON(offset_in_page(start));
522 VM_BUG_ON(len != PAGE_ALIGN(len));
523 end = start + len;
524 if (end < start)
525 return -EINVAL;
526 if (end == start)
527 return 0;
528 vma = vma_iter_load(&vmi);
529 if (!vma)
530 return -ENOMEM;
531
532 prev = vma_prev(&vmi);
533 if (start > vma->vm_start)
534 prev = vma;
535
536 nstart = start;
537 tmp = vma->vm_start;
538 for_each_vma_range(vmi, vma, end) {
539 int error;
540 vm_flags_t newflags;
541
542 if (vma->vm_start != tmp)
543 return -ENOMEM;
544
545 newflags = vma->vm_flags & ~VM_LOCKED_MASK;
546 newflags |= flags;
547 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
548 tmp = vma->vm_end;
549 if (tmp > end)
550 tmp = end;
551 error = mlock_fixup(&vmi, vma, &prev, nstart, tmp, newflags);
552 if (error)
553 return error;
554 tmp = vma_iter_end(&vmi);
555 nstart = tmp;
556 }
557
558 if (tmp < end)
559 return -ENOMEM;
560
561 return 0;
562}
563
564/*
565 * Go through vma areas and sum size of mlocked
566 * vma pages, as return value.
567 * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT)
568 * is also counted.
569 * Return value: previously mlocked page counts
570 */
571static unsigned long count_mm_mlocked_page_nr(struct mm_struct *mm,
572 unsigned long start, size_t len)
573{
574 struct vm_area_struct *vma;
575 unsigned long count = 0;
576 unsigned long end;
577 VMA_ITERATOR(vmi, mm, start);
578
579 /* Don't overflow past ULONG_MAX */
580 if (unlikely(ULONG_MAX - len < start))
581 end = ULONG_MAX;
582 else
583 end = start + len;
584
585 for_each_vma_range(vmi, vma, end) {
586 if (vma->vm_flags & VM_LOCKED) {
587 if (start > vma->vm_start)
588 count -= (start - vma->vm_start);
589 if (end < vma->vm_end) {
590 count += end - vma->vm_start;
591 break;
592 }
593 count += vma->vm_end - vma->vm_start;
594 }
595 }
596
597 return count >> PAGE_SHIFT;
598}
599
600/*
601 * convert get_user_pages() return value to posix mlock() error
602 */
603static int __mlock_posix_error_return(long retval)
604{
605 if (retval == -EFAULT)
606 retval = -ENOMEM;
607 else if (retval == -ENOMEM)
608 retval = -EAGAIN;
609 return retval;
610}
611
612static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
613{
614 unsigned long locked;
615 unsigned long lock_limit;
616 int error = -ENOMEM;
617
618 start = untagged_addr(start);
619
620 if (!can_do_mlock())
621 return -EPERM;
622
623 len = PAGE_ALIGN(len + (offset_in_page(start)));
624 start &= PAGE_MASK;
625
626 lock_limit = rlimit(RLIMIT_MEMLOCK);
627 lock_limit >>= PAGE_SHIFT;
628 locked = len >> PAGE_SHIFT;
629
630 if (mmap_write_lock_killable(current->mm))
631 return -EINTR;
632
633 locked += current->mm->locked_vm;
634 if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) {
635 /*
636 * It is possible that the regions requested intersect with
637 * previously mlocked areas, that part area in "mm->locked_vm"
638 * should not be counted to new mlock increment count. So check
639 * and adjust locked count if necessary.
640 */
641 locked -= count_mm_mlocked_page_nr(current->mm,
642 start, len);
643 }
644
645 /* check against resource limits */
646 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
647 error = apply_vma_lock_flags(start, len, flags);
648
649 mmap_write_unlock(current->mm);
650 if (error)
651 return error;
652
653 error = __mm_populate(start, len, 0);
654 if (error)
655 return __mlock_posix_error_return(error);
656 return 0;
657}
658
659SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
660{
661 return do_mlock(start, len, VM_LOCKED);
662}
663
664SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
665{
666 vm_flags_t vm_flags = VM_LOCKED;
667
668 if (flags & ~MLOCK_ONFAULT)
669 return -EINVAL;
670
671 if (flags & MLOCK_ONFAULT)
672 vm_flags |= VM_LOCKONFAULT;
673
674 return do_mlock(start, len, vm_flags);
675}
676
677SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
678{
679 int ret;
680
681 start = untagged_addr(start);
682
683 len = PAGE_ALIGN(len + (offset_in_page(start)));
684 start &= PAGE_MASK;
685
686 if (mmap_write_lock_killable(current->mm))
687 return -EINTR;
688 ret = apply_vma_lock_flags(start, len, 0);
689 mmap_write_unlock(current->mm);
690
691 return ret;
692}
693
694/*
695 * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
696 * and translate into the appropriate modifications to mm->def_flags and/or the
697 * flags for all current VMAs.
698 *
699 * There are a couple of subtleties with this. If mlockall() is called multiple
700 * times with different flags, the values do not necessarily stack. If mlockall
701 * is called once including the MCL_FUTURE flag and then a second time without
702 * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
703 */
704static int apply_mlockall_flags(int flags)
705{
706 VMA_ITERATOR(vmi, current->mm, 0);
707 struct vm_area_struct *vma, *prev = NULL;
708 vm_flags_t to_add = 0;
709
710 current->mm->def_flags &= ~VM_LOCKED_MASK;
711 if (flags & MCL_FUTURE) {
712 current->mm->def_flags |= VM_LOCKED;
713
714 if (flags & MCL_ONFAULT)
715 current->mm->def_flags |= VM_LOCKONFAULT;
716
717 if (!(flags & MCL_CURRENT))
718 goto out;
719 }
720
721 if (flags & MCL_CURRENT) {
722 to_add |= VM_LOCKED;
723 if (flags & MCL_ONFAULT)
724 to_add |= VM_LOCKONFAULT;
725 }
726
727 for_each_vma(vmi, vma) {
728 int error;
729 vm_flags_t newflags;
730
731 newflags = vma->vm_flags & ~VM_LOCKED_MASK;
732 newflags |= to_add;
733
734 error = mlock_fixup(&vmi, vma, &prev, vma->vm_start, vma->vm_end,
735 newflags);
736 /* Ignore errors, but prev needs fixing up. */
737 if (error)
738 prev = vma;
739 cond_resched();
740 }
741out:
742 return 0;
743}
744
745SYSCALL_DEFINE1(mlockall, int, flags)
746{
747 unsigned long lock_limit;
748 int ret;
749
750 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)) ||
751 flags == MCL_ONFAULT)
752 return -EINVAL;
753
754 if (!can_do_mlock())
755 return -EPERM;
756
757 lock_limit = rlimit(RLIMIT_MEMLOCK);
758 lock_limit >>= PAGE_SHIFT;
759
760 if (mmap_write_lock_killable(current->mm))
761 return -EINTR;
762
763 ret = -ENOMEM;
764 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
765 capable(CAP_IPC_LOCK))
766 ret = apply_mlockall_flags(flags);
767 mmap_write_unlock(current->mm);
768 if (!ret && (flags & MCL_CURRENT))
769 mm_populate(0, TASK_SIZE);
770
771 return ret;
772}
773
774SYSCALL_DEFINE0(munlockall)
775{
776 int ret;
777
778 if (mmap_write_lock_killable(current->mm))
779 return -EINTR;
780 ret = apply_mlockall_flags(0);
781 mmap_write_unlock(current->mm);
782 return ret;
783}
784
785/*
786 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
787 * shm segments) get accounted against the user_struct instead.
788 */
789static DEFINE_SPINLOCK(shmlock_user_lock);
790
791int user_shm_lock(size_t size, struct ucounts *ucounts)
792{
793 unsigned long lock_limit, locked;
794 long memlock;
795 int allowed = 0;
796
797 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
798 lock_limit = rlimit(RLIMIT_MEMLOCK);
799 if (lock_limit != RLIM_INFINITY)
800 lock_limit >>= PAGE_SHIFT;
801 spin_lock(&shmlock_user_lock);
802 memlock = inc_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
803
804 if ((memlock == LONG_MAX || memlock > lock_limit) && !capable(CAP_IPC_LOCK)) {
805 dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
806 goto out;
807 }
808 if (!get_ucounts(ucounts)) {
809 dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
810 allowed = 0;
811 goto out;
812 }
813 allowed = 1;
814out:
815 spin_unlock(&shmlock_user_lock);
816 return allowed;
817}
818
819void user_shm_unlock(size_t size, struct ucounts *ucounts)
820{
821 spin_lock(&shmlock_user_lock);
822 dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, (size + PAGE_SIZE - 1) >> PAGE_SHIFT);
823 spin_unlock(&shmlock_user_lock);
824 put_ucounts(ucounts);
825}
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_pgdat(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(zone));
192
193 if (!TestClearPageMlocked(page)) {
194 /* Potentially, PTE-mapped THP: do not skip the rest PTEs */
195 nr_pages = 1;
196 goto unlock_out;
197 }
198
199 nr_pages = hpage_nr_pages(page);
200 __mod_zone_page_state(zone, NR_MLOCK, -nr_pages);
201
202 if (__munlock_isolate_lru_page(page, true)) {
203 spin_unlock_irq(zone_lru_lock(zone));
204 __munlock_isolated_page(page);
205 goto out;
206 }
207 __munlock_isolation_failed(page);
208
209unlock_out:
210 spin_unlock_irq(zone_lru_lock(zone));
211
212out:
213 return nr_pages - 1;
214}
215
216/*
217 * convert get_user_pages() return value to posix mlock() error
218 */
219static int __mlock_posix_error_return(long retval)
220{
221 if (retval == -EFAULT)
222 retval = -ENOMEM;
223 else if (retval == -ENOMEM)
224 retval = -EAGAIN;
225 return retval;
226}
227
228/*
229 * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec()
230 *
231 * The fast path is available only for evictable pages with single mapping.
232 * Then we can bypass the per-cpu pvec and get better performance.
233 * when mapcount > 1 we need try_to_munlock() which can fail.
234 * when !page_evictable(), we need the full redo logic of putback_lru_page to
235 * avoid leaving evictable page in unevictable list.
236 *
237 * In case of success, @page is added to @pvec and @pgrescued is incremented
238 * in case that the page was previously unevictable. @page is also unlocked.
239 */
240static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec,
241 int *pgrescued)
242{
243 VM_BUG_ON_PAGE(PageLRU(page), page);
244 VM_BUG_ON_PAGE(!PageLocked(page), page);
245
246 if (page_mapcount(page) <= 1 && page_evictable(page)) {
247 pagevec_add(pvec, page);
248 if (TestClearPageUnevictable(page))
249 (*pgrescued)++;
250 unlock_page(page);
251 return true;
252 }
253
254 return false;
255}
256
257/*
258 * Putback multiple evictable pages to the LRU
259 *
260 * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of
261 * the pages might have meanwhile become unevictable but that is OK.
262 */
263static void __putback_lru_fast(struct pagevec *pvec, int pgrescued)
264{
265 count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec));
266 /*
267 *__pagevec_lru_add() calls release_pages() so we don't call
268 * put_page() explicitly
269 */
270 __pagevec_lru_add(pvec);
271 count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued);
272}
273
274/*
275 * Munlock a batch of pages from the same zone
276 *
277 * The work is split to two main phases. First phase clears the Mlocked flag
278 * and attempts to isolate the pages, all under a single zone lru lock.
279 * The second phase finishes the munlock only for pages where isolation
280 * succeeded.
281 *
282 * Note that the pagevec may be modified during the process.
283 */
284static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone)
285{
286 int i;
287 int nr = pagevec_count(pvec);
288 int delta_munlocked;
289 struct pagevec pvec_putback;
290 int pgrescued = 0;
291
292 pagevec_init(&pvec_putback, 0);
293
294 /* Phase 1: page isolation */
295 spin_lock_irq(zone_lru_lock(zone));
296 for (i = 0; i < nr; i++) {
297 struct page *page = pvec->pages[i];
298
299 if (TestClearPageMlocked(page)) {
300 /*
301 * We already have pin from follow_page_mask()
302 * so we can spare the get_page() here.
303 */
304 if (__munlock_isolate_lru_page(page, false))
305 continue;
306 else
307 __munlock_isolation_failed(page);
308 }
309
310 /*
311 * We won't be munlocking this page in the next phase
312 * but we still need to release the follow_page_mask()
313 * pin. We cannot do it under lru_lock however. If it's
314 * the last pin, __page_cache_release() would deadlock.
315 */
316 pagevec_add(&pvec_putback, pvec->pages[i]);
317 pvec->pages[i] = NULL;
318 }
319 delta_munlocked = -nr + pagevec_count(&pvec_putback);
320 __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked);
321 spin_unlock_irq(zone_lru_lock(zone));
322
323 /* Now we can release pins of pages that we are not munlocking */
324 pagevec_release(&pvec_putback);
325
326 /* Phase 2: page munlock */
327 for (i = 0; i < nr; i++) {
328 struct page *page = pvec->pages[i];
329
330 if (page) {
331 lock_page(page);
332 if (!__putback_lru_fast_prepare(page, &pvec_putback,
333 &pgrescued)) {
334 /*
335 * Slow path. We don't want to lose the last
336 * pin before unlock_page()
337 */
338 get_page(page); /* for putback_lru_page() */
339 __munlock_isolated_page(page);
340 unlock_page(page);
341 put_page(page); /* from follow_page_mask() */
342 }
343 }
344 }
345
346 /*
347 * Phase 3: page putback for pages that qualified for the fast path
348 * This will also call put_page() to return pin from follow_page_mask()
349 */
350 if (pagevec_count(&pvec_putback))
351 __putback_lru_fast(&pvec_putback, pgrescued);
352}
353
354/*
355 * Fill up pagevec for __munlock_pagevec using pte walk
356 *
357 * The function expects that the struct page corresponding to @start address is
358 * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone.
359 *
360 * The rest of @pvec is filled by subsequent pages within the same pmd and same
361 * zone, as long as the pte's are present and vm_normal_page() succeeds. These
362 * pages also get pinned.
363 *
364 * Returns the address of the next page that should be scanned. This equals
365 * @start + PAGE_SIZE when no page could be added by the pte walk.
366 */
367static unsigned long __munlock_pagevec_fill(struct pagevec *pvec,
368 struct vm_area_struct *vma, int zoneid, unsigned long start,
369 unsigned long end)
370{
371 pte_t *pte;
372 spinlock_t *ptl;
373
374 /*
375 * Initialize pte walk starting at the already pinned page where we
376 * are sure that there is a pte, as it was pinned under the same
377 * mmap_sem write op.
378 */
379 pte = get_locked_pte(vma->vm_mm, start, &ptl);
380 /* Make sure we do not cross the page table boundary */
381 end = pgd_addr_end(start, end);
382 end = pud_addr_end(start, end);
383 end = pmd_addr_end(start, end);
384
385 /* The page next to the pinned page is the first we will try to get */
386 start += PAGE_SIZE;
387 while (start < end) {
388 struct page *page = NULL;
389 pte++;
390 if (pte_present(*pte))
391 page = vm_normal_page(vma, start, *pte);
392 /*
393 * Break if page could not be obtained or the page's node+zone does not
394 * match
395 */
396 if (!page || page_zone_id(page) != zoneid)
397 break;
398
399 /*
400 * Do not use pagevec for PTE-mapped THP,
401 * munlock_vma_pages_range() will handle them.
402 */
403 if (PageTransCompound(page))
404 break;
405
406 get_page(page);
407 /*
408 * Increase the address that will be returned *before* the
409 * eventual break due to pvec becoming full by adding the page
410 */
411 start += PAGE_SIZE;
412 if (pagevec_add(pvec, page) == 0)
413 break;
414 }
415 pte_unmap_unlock(pte, ptl);
416 return start;
417}
418
419/*
420 * munlock_vma_pages_range() - munlock all pages in the vma range.'
421 * @vma - vma containing range to be munlock()ed.
422 * @start - start address in @vma of the range
423 * @end - end of range in @vma.
424 *
425 * For mremap(), munmap() and exit().
426 *
427 * Called with @vma VM_LOCKED.
428 *
429 * Returns with VM_LOCKED cleared. Callers must be prepared to
430 * deal with this.
431 *
432 * We don't save and restore VM_LOCKED here because pages are
433 * still on lru. In unmap path, pages might be scanned by reclaim
434 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
435 * free them. This will result in freeing mlocked pages.
436 */
437void munlock_vma_pages_range(struct vm_area_struct *vma,
438 unsigned long start, unsigned long end)
439{
440 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
441
442 while (start < end) {
443 struct page *page;
444 unsigned int page_mask = 0;
445 unsigned long page_increm;
446 struct pagevec pvec;
447 struct zone *zone;
448 int zoneid;
449
450 pagevec_init(&pvec, 0);
451 /*
452 * Although FOLL_DUMP is intended for get_dump_page(),
453 * it just so happens that its special treatment of the
454 * ZERO_PAGE (returning an error instead of doing get_page)
455 * suits munlock very well (and if somehow an abnormal page
456 * has sneaked into the range, we won't oops here: great).
457 */
458 page = follow_page(vma, start, FOLL_GET | FOLL_DUMP);
459
460 if (page && !IS_ERR(page)) {
461 if (PageTransTail(page)) {
462 VM_BUG_ON_PAGE(PageMlocked(page), page);
463 put_page(page); /* follow_page_mask() */
464 } else if (PageTransHuge(page)) {
465 lock_page(page);
466 /*
467 * Any THP page found by follow_page_mask() may
468 * have gotten split before reaching
469 * munlock_vma_page(), so we need to compute
470 * the page_mask here instead.
471 */
472 page_mask = munlock_vma_page(page);
473 unlock_page(page);
474 put_page(page); /* follow_page_mask() */
475 } else {
476 /*
477 * Non-huge pages are handled in batches via
478 * pagevec. The pin from follow_page_mask()
479 * prevents them from collapsing by THP.
480 */
481 pagevec_add(&pvec, page);
482 zone = page_zone(page);
483 zoneid = page_zone_id(page);
484
485 /*
486 * Try to fill the rest of pagevec using fast
487 * pte walk. This will also update start to
488 * the next page to process. Then munlock the
489 * pagevec.
490 */
491 start = __munlock_pagevec_fill(&pvec, vma,
492 zoneid, start, end);
493 __munlock_pagevec(&pvec, zone);
494 goto next;
495 }
496 }
497 page_increm = 1 + page_mask;
498 start += page_increm * PAGE_SIZE;
499next:
500 cond_resched();
501 }
502}
503
504/*
505 * mlock_fixup - handle mlock[all]/munlock[all] requests.
506 *
507 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
508 * munlock is a no-op. However, for some special vmas, we go ahead and
509 * populate the ptes.
510 *
511 * For vmas that pass the filters, merge/split as appropriate.
512 */
513static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
514 unsigned long start, unsigned long end, vm_flags_t newflags)
515{
516 struct mm_struct *mm = vma->vm_mm;
517 pgoff_t pgoff;
518 int nr_pages;
519 int ret = 0;
520 int lock = !!(newflags & VM_LOCKED);
521 vm_flags_t old_flags = vma->vm_flags;
522
523 if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
524 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
525 /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
526 goto out;
527
528 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
529 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
530 vma->vm_file, pgoff, vma_policy(vma),
531 vma->vm_userfaultfd_ctx);
532 if (*prev) {
533 vma = *prev;
534 goto success;
535 }
536
537 if (start != vma->vm_start) {
538 ret = split_vma(mm, vma, start, 1);
539 if (ret)
540 goto out;
541 }
542
543 if (end != vma->vm_end) {
544 ret = split_vma(mm, vma, end, 0);
545 if (ret)
546 goto out;
547 }
548
549success:
550 /*
551 * Keep track of amount of locked VM.
552 */
553 nr_pages = (end - start) >> PAGE_SHIFT;
554 if (!lock)
555 nr_pages = -nr_pages;
556 else if (old_flags & VM_LOCKED)
557 nr_pages = 0;
558 mm->locked_vm += nr_pages;
559
560 /*
561 * vm_flags is protected by the mmap_sem held in write mode.
562 * It's okay if try_to_unmap_one unmaps a page just after we
563 * set VM_LOCKED, populate_vma_page_range will bring it back.
564 */
565
566 if (lock)
567 vma->vm_flags = newflags;
568 else
569 munlock_vma_pages_range(vma, start, end);
570
571out:
572 *prev = vma;
573 return ret;
574}
575
576static int apply_vma_lock_flags(unsigned long start, size_t len,
577 vm_flags_t flags)
578{
579 unsigned long nstart, end, tmp;
580 struct vm_area_struct * vma, * prev;
581 int error;
582
583 VM_BUG_ON(offset_in_page(start));
584 VM_BUG_ON(len != PAGE_ALIGN(len));
585 end = start + len;
586 if (end < start)
587 return -EINVAL;
588 if (end == start)
589 return 0;
590 vma = find_vma(current->mm, start);
591 if (!vma || vma->vm_start > start)
592 return -ENOMEM;
593
594 prev = vma->vm_prev;
595 if (start > vma->vm_start)
596 prev = vma;
597
598 for (nstart = start ; ; ) {
599 vm_flags_t newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
600
601 newflags |= flags;
602
603 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
604 tmp = vma->vm_end;
605 if (tmp > end)
606 tmp = end;
607 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
608 if (error)
609 break;
610 nstart = tmp;
611 if (nstart < prev->vm_end)
612 nstart = prev->vm_end;
613 if (nstart >= end)
614 break;
615
616 vma = prev->vm_next;
617 if (!vma || vma->vm_start != nstart) {
618 error = -ENOMEM;
619 break;
620 }
621 }
622 return error;
623}
624
625/*
626 * Go through vma areas and sum size of mlocked
627 * vma pages, as return value.
628 * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT)
629 * is also counted.
630 * Return value: previously mlocked page counts
631 */
632static int count_mm_mlocked_page_nr(struct mm_struct *mm,
633 unsigned long start, size_t len)
634{
635 struct vm_area_struct *vma;
636 int count = 0;
637
638 if (mm == NULL)
639 mm = current->mm;
640
641 vma = find_vma(mm, start);
642 if (vma == NULL)
643 vma = mm->mmap;
644
645 for (; vma ; vma = vma->vm_next) {
646 if (start >= vma->vm_end)
647 continue;
648 if (start + len <= vma->vm_start)
649 break;
650 if (vma->vm_flags & VM_LOCKED) {
651 if (start > vma->vm_start)
652 count -= (start - vma->vm_start);
653 if (start + len < vma->vm_end) {
654 count += start + len - vma->vm_start;
655 break;
656 }
657 count += vma->vm_end - vma->vm_start;
658 }
659 }
660
661 return count >> PAGE_SHIFT;
662}
663
664static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
665{
666 unsigned long locked;
667 unsigned long lock_limit;
668 int error = -ENOMEM;
669
670 if (!can_do_mlock())
671 return -EPERM;
672
673 lru_add_drain_all(); /* flush pagevec */
674
675 len = PAGE_ALIGN(len + (offset_in_page(start)));
676 start &= PAGE_MASK;
677
678 lock_limit = rlimit(RLIMIT_MEMLOCK);
679 lock_limit >>= PAGE_SHIFT;
680 locked = len >> PAGE_SHIFT;
681
682 if (down_write_killable(¤t->mm->mmap_sem))
683 return -EINTR;
684
685 locked += current->mm->locked_vm;
686 if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) {
687 /*
688 * It is possible that the regions requested intersect with
689 * previously mlocked areas, that part area in "mm->locked_vm"
690 * should not be counted to new mlock increment count. So check
691 * and adjust locked count if necessary.
692 */
693 locked -= count_mm_mlocked_page_nr(current->mm,
694 start, len);
695 }
696
697 /* check against resource limits */
698 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
699 error = apply_vma_lock_flags(start, len, flags);
700
701 up_write(¤t->mm->mmap_sem);
702 if (error)
703 return error;
704
705 error = __mm_populate(start, len, 0);
706 if (error)
707 return __mlock_posix_error_return(error);
708 return 0;
709}
710
711SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
712{
713 return do_mlock(start, len, VM_LOCKED);
714}
715
716SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
717{
718 vm_flags_t vm_flags = VM_LOCKED;
719
720 if (flags & ~MLOCK_ONFAULT)
721 return -EINVAL;
722
723 if (flags & MLOCK_ONFAULT)
724 vm_flags |= VM_LOCKONFAULT;
725
726 return do_mlock(start, len, vm_flags);
727}
728
729SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
730{
731 int ret;
732
733 len = PAGE_ALIGN(len + (offset_in_page(start)));
734 start &= PAGE_MASK;
735
736 if (down_write_killable(¤t->mm->mmap_sem))
737 return -EINTR;
738 ret = apply_vma_lock_flags(start, len, 0);
739 up_write(¤t->mm->mmap_sem);
740
741 return ret;
742}
743
744/*
745 * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
746 * and translate into the appropriate modifications to mm->def_flags and/or the
747 * flags for all current VMAs.
748 *
749 * There are a couple of subtleties with this. If mlockall() is called multiple
750 * times with different flags, the values do not necessarily stack. If mlockall
751 * is called once including the MCL_FUTURE flag and then a second time without
752 * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
753 */
754static int apply_mlockall_flags(int flags)
755{
756 struct vm_area_struct * vma, * prev = NULL;
757 vm_flags_t to_add = 0;
758
759 current->mm->def_flags &= VM_LOCKED_CLEAR_MASK;
760 if (flags & MCL_FUTURE) {
761 current->mm->def_flags |= VM_LOCKED;
762
763 if (flags & MCL_ONFAULT)
764 current->mm->def_flags |= VM_LOCKONFAULT;
765
766 if (!(flags & MCL_CURRENT))
767 goto out;
768 }
769
770 if (flags & MCL_CURRENT) {
771 to_add |= VM_LOCKED;
772 if (flags & MCL_ONFAULT)
773 to_add |= VM_LOCKONFAULT;
774 }
775
776 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
777 vm_flags_t newflags;
778
779 newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
780 newflags |= to_add;
781
782 /* Ignore errors */
783 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
784 cond_resched_rcu_qs();
785 }
786out:
787 return 0;
788}
789
790SYSCALL_DEFINE1(mlockall, int, flags)
791{
792 unsigned long lock_limit;
793 int ret;
794
795 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)))
796 return -EINVAL;
797
798 if (!can_do_mlock())
799 return -EPERM;
800
801 if (flags & MCL_CURRENT)
802 lru_add_drain_all(); /* flush pagevec */
803
804 lock_limit = rlimit(RLIMIT_MEMLOCK);
805 lock_limit >>= PAGE_SHIFT;
806
807 if (down_write_killable(¤t->mm->mmap_sem))
808 return -EINTR;
809
810 ret = -ENOMEM;
811 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
812 capable(CAP_IPC_LOCK))
813 ret = apply_mlockall_flags(flags);
814 up_write(¤t->mm->mmap_sem);
815 if (!ret && (flags & MCL_CURRENT))
816 mm_populate(0, TASK_SIZE);
817
818 return ret;
819}
820
821SYSCALL_DEFINE0(munlockall)
822{
823 int ret;
824
825 if (down_write_killable(¤t->mm->mmap_sem))
826 return -EINTR;
827 ret = apply_mlockall_flags(0);
828 up_write(¤t->mm->mmap_sem);
829 return ret;
830}
831
832/*
833 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
834 * shm segments) get accounted against the user_struct instead.
835 */
836static DEFINE_SPINLOCK(shmlock_user_lock);
837
838int user_shm_lock(size_t size, struct user_struct *user)
839{
840 unsigned long lock_limit, locked;
841 int allowed = 0;
842
843 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
844 lock_limit = rlimit(RLIMIT_MEMLOCK);
845 if (lock_limit == RLIM_INFINITY)
846 allowed = 1;
847 lock_limit >>= PAGE_SHIFT;
848 spin_lock(&shmlock_user_lock);
849 if (!allowed &&
850 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
851 goto out;
852 get_uid(user);
853 user->locked_shm += locked;
854 allowed = 1;
855out:
856 spin_unlock(&shmlock_user_lock);
857 return allowed;
858}
859
860void user_shm_unlock(size_t size, struct user_struct *user)
861{
862 spin_lock(&shmlock_user_lock);
863 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
864 spin_unlock(&shmlock_user_lock);
865 free_uid(user);
866}