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// 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_pvec {
32 local_lock_t lock;
33 struct pagevec vec;
34};
35
36static DEFINE_PER_CPU(struct mlock_pvec, mlock_pvec) = {
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 pages are marked with PageMlocked() flag for efficient testing
52 * in vmscan and, possibly, the fault path; and to support semi-accurate
53 * statistics.
54 *
55 * An mlocked page [PageMlocked(page)] is unevictable. As such, it will
56 * be placed on the LRU "unevictable" list, rather than the [in]active lists.
57 * The unevictable list is an LRU sibling list to the [in]active lists.
58 * PageUnevictable is set to indicate the unevictable state.
59 */
60
61static struct lruvec *__mlock_page(struct page *page, struct lruvec *lruvec)
62{
63 /* There is nothing more we can do while it's off LRU */
64 if (!TestClearPageLRU(page))
65 return lruvec;
66
67 lruvec = folio_lruvec_relock_irq(page_folio(page), lruvec);
68
69 if (unlikely(page_evictable(page))) {
70 /*
71 * This is a little surprising, but quite possible:
72 * PageMlocked must have got cleared already by another CPU.
73 * Could this page be on the Unevictable LRU? I'm not sure,
74 * but move it now if so.
75 */
76 if (PageUnevictable(page)) {
77 del_page_from_lru_list(page, lruvec);
78 ClearPageUnevictable(page);
79 add_page_to_lru_list(page, lruvec);
80 __count_vm_events(UNEVICTABLE_PGRESCUED,
81 thp_nr_pages(page));
82 }
83 goto out;
84 }
85
86 if (PageUnevictable(page)) {
87 if (PageMlocked(page))
88 page->mlock_count++;
89 goto out;
90 }
91
92 del_page_from_lru_list(page, lruvec);
93 ClearPageActive(page);
94 SetPageUnevictable(page);
95 page->mlock_count = !!PageMlocked(page);
96 add_page_to_lru_list(page, lruvec);
97 __count_vm_events(UNEVICTABLE_PGCULLED, thp_nr_pages(page));
98out:
99 SetPageLRU(page);
100 return lruvec;
101}
102
103static struct lruvec *__mlock_new_page(struct page *page, struct lruvec *lruvec)
104{
105 VM_BUG_ON_PAGE(PageLRU(page), page);
106
107 lruvec = folio_lruvec_relock_irq(page_folio(page), lruvec);
108
109 /* As above, this is a little surprising, but possible */
110 if (unlikely(page_evictable(page)))
111 goto out;
112
113 SetPageUnevictable(page);
114 page->mlock_count = !!PageMlocked(page);
115 __count_vm_events(UNEVICTABLE_PGCULLED, thp_nr_pages(page));
116out:
117 add_page_to_lru_list(page, lruvec);
118 SetPageLRU(page);
119 return lruvec;
120}
121
122static struct lruvec *__munlock_page(struct page *page, struct lruvec *lruvec)
123{
124 int nr_pages = thp_nr_pages(page);
125 bool isolated = false;
126
127 if (!TestClearPageLRU(page))
128 goto munlock;
129
130 isolated = true;
131 lruvec = folio_lruvec_relock_irq(page_folio(page), lruvec);
132
133 if (PageUnevictable(page)) {
134 /* Then mlock_count is maintained, but might undercount */
135 if (page->mlock_count)
136 page->mlock_count--;
137 if (page->mlock_count)
138 goto out;
139 }
140 /* else assume that was the last mlock: reclaim will fix it if not */
141
142munlock:
143 if (TestClearPageMlocked(page)) {
144 __mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
145 if (isolated || !PageUnevictable(page))
146 __count_vm_events(UNEVICTABLE_PGMUNLOCKED, nr_pages);
147 else
148 __count_vm_events(UNEVICTABLE_PGSTRANDED, nr_pages);
149 }
150
151 /* page_evictable() has to be checked *after* clearing Mlocked */
152 if (isolated && PageUnevictable(page) && page_evictable(page)) {
153 del_page_from_lru_list(page, lruvec);
154 ClearPageUnevictable(page);
155 add_page_to_lru_list(page, lruvec);
156 __count_vm_events(UNEVICTABLE_PGRESCUED, nr_pages);
157 }
158out:
159 if (isolated)
160 SetPageLRU(page);
161 return lruvec;
162}
163
164/*
165 * Flags held in the low bits of a struct page pointer on the mlock_pvec.
166 */
167#define LRU_PAGE 0x1
168#define NEW_PAGE 0x2
169static inline struct page *mlock_lru(struct page *page)
170{
171 return (struct page *)((unsigned long)page + LRU_PAGE);
172}
173
174static inline struct page *mlock_new(struct page *page)
175{
176 return (struct page *)((unsigned long)page + NEW_PAGE);
177}
178
179/*
180 * mlock_pagevec() is derived from pagevec_lru_move_fn():
181 * perhaps that can make use of such page pointer flags in future,
182 * but for now just keep it for mlock. We could use three separate
183 * pagevecs instead, but one feels better (munlocking a full pagevec
184 * does not need to drain mlocking pagevecs first).
185 */
186static void mlock_pagevec(struct pagevec *pvec)
187{
188 struct lruvec *lruvec = NULL;
189 unsigned long mlock;
190 struct page *page;
191 int i;
192
193 for (i = 0; i < pagevec_count(pvec); i++) {
194 page = pvec->pages[i];
195 mlock = (unsigned long)page & (LRU_PAGE | NEW_PAGE);
196 page = (struct page *)((unsigned long)page - mlock);
197 pvec->pages[i] = page;
198
199 if (mlock & LRU_PAGE)
200 lruvec = __mlock_page(page, lruvec);
201 else if (mlock & NEW_PAGE)
202 lruvec = __mlock_new_page(page, lruvec);
203 else
204 lruvec = __munlock_page(page, lruvec);
205 }
206
207 if (lruvec)
208 unlock_page_lruvec_irq(lruvec);
209 release_pages(pvec->pages, pvec->nr);
210 pagevec_reinit(pvec);
211}
212
213void mlock_page_drain_local(void)
214{
215 struct pagevec *pvec;
216
217 local_lock(&mlock_pvec.lock);
218 pvec = this_cpu_ptr(&mlock_pvec.vec);
219 if (pagevec_count(pvec))
220 mlock_pagevec(pvec);
221 local_unlock(&mlock_pvec.lock);
222}
223
224void mlock_page_drain_remote(int cpu)
225{
226 struct pagevec *pvec;
227
228 WARN_ON_ONCE(cpu_online(cpu));
229 pvec = &per_cpu(mlock_pvec.vec, cpu);
230 if (pagevec_count(pvec))
231 mlock_pagevec(pvec);
232}
233
234bool need_mlock_page_drain(int cpu)
235{
236 return pagevec_count(&per_cpu(mlock_pvec.vec, 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 pagevec *pvec;
246
247 local_lock(&mlock_pvec.lock);
248 pvec = this_cpu_ptr(&mlock_pvec.vec);
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 (!pagevec_add(pvec, mlock_lru(&folio->page)) ||
259 folio_test_large(folio) || lru_cache_disabled())
260 mlock_pagevec(pvec);
261 local_unlock(&mlock_pvec.lock);
262}
263
264/**
265 * mlock_new_page - mlock a newly allocated page not yet on LRU
266 * @page: page to be mlocked, either a normal page or a THP head.
267 */
268void mlock_new_page(struct page *page)
269{
270 struct pagevec *pvec;
271 int nr_pages = thp_nr_pages(page);
272
273 local_lock(&mlock_pvec.lock);
274 pvec = this_cpu_ptr(&mlock_pvec.vec);
275 SetPageMlocked(page);
276 mod_zone_page_state(page_zone(page), NR_MLOCK, nr_pages);
277 __count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
278
279 get_page(page);
280 if (!pagevec_add(pvec, mlock_new(page)) ||
281 PageHead(page) || lru_cache_disabled())
282 mlock_pagevec(pvec);
283 local_unlock(&mlock_pvec.lock);
284}
285
286/**
287 * munlock_page - munlock a page
288 * @page: page to be munlocked, either a normal page or a THP head.
289 */
290void munlock_page(struct page *page)
291{
292 struct pagevec *pvec;
293
294 local_lock(&mlock_pvec.lock);
295 pvec = this_cpu_ptr(&mlock_pvec.vec);
296 /*
297 * TestClearPageMlocked(page) must be left to __munlock_page(),
298 * which will check whether the page is multiply mlocked.
299 */
300
301 get_page(page);
302 if (!pagevec_add(pvec, page) ||
303 PageHead(page) || lru_cache_disabled())
304 mlock_pagevec(pvec);
305 local_unlock(&mlock_pvec.lock);
306}
307
308static int mlock_pte_range(pmd_t *pmd, unsigned long addr,
309 unsigned long end, struct mm_walk *walk)
310
311{
312 struct vm_area_struct *vma = walk->vma;
313 spinlock_t *ptl;
314 pte_t *start_pte, *pte;
315 struct page *page;
316
317 ptl = pmd_trans_huge_lock(pmd, vma);
318 if (ptl) {
319 if (!pmd_present(*pmd))
320 goto out;
321 if (is_huge_zero_pmd(*pmd))
322 goto out;
323 page = pmd_page(*pmd);
324 if (vma->vm_flags & VM_LOCKED)
325 mlock_folio(page_folio(page));
326 else
327 munlock_page(page);
328 goto out;
329 }
330
331 start_pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
332 for (pte = start_pte; addr != end; pte++, addr += PAGE_SIZE) {
333 if (!pte_present(*pte))
334 continue;
335 page = vm_normal_page(vma, addr, *pte);
336 if (!page || is_zone_device_page(page))
337 continue;
338 if (PageTransCompound(page))
339 continue;
340 if (vma->vm_flags & VM_LOCKED)
341 mlock_folio(page_folio(page));
342 else
343 munlock_page(page);
344 }
345 pte_unmap(start_pte);
346out:
347 spin_unlock(ptl);
348 cond_resched();
349 return 0;
350}
351
352/*
353 * mlock_vma_pages_range() - mlock any pages already in the range,
354 * or munlock all pages in the range.
355 * @vma - vma containing range to be mlock()ed or munlock()ed
356 * @start - start address in @vma of the range
357 * @end - end of range in @vma
358 * @newflags - the new set of flags for @vma.
359 *
360 * Called for mlock(), mlock2() and mlockall(), to set @vma VM_LOCKED;
361 * called for munlock() and munlockall(), to clear VM_LOCKED from @vma.
362 */
363static void mlock_vma_pages_range(struct vm_area_struct *vma,
364 unsigned long start, unsigned long end, vm_flags_t newflags)
365{
366 static const struct mm_walk_ops mlock_walk_ops = {
367 .pmd_entry = mlock_pte_range,
368 };
369
370 /*
371 * There is a slight chance that concurrent page migration,
372 * or page reclaim finding a page of this now-VM_LOCKED vma,
373 * will call mlock_vma_page() and raise page's mlock_count:
374 * double counting, leaving the page unevictable indefinitely.
375 * Communicate this danger to mlock_vma_page() with VM_IO,
376 * which is a VM_SPECIAL flag not allowed on VM_LOCKED vmas.
377 * mmap_lock is held in write mode here, so this weird
378 * combination should not be visible to other mmap_lock users;
379 * but WRITE_ONCE so rmap walkers must see VM_IO if VM_LOCKED.
380 */
381 if (newflags & VM_LOCKED)
382 newflags |= VM_IO;
383 WRITE_ONCE(vma->vm_flags, newflags);
384
385 lru_add_drain();
386 walk_page_range(vma->vm_mm, start, end, &mlock_walk_ops, NULL);
387 lru_add_drain();
388
389 if (newflags & VM_IO) {
390 newflags &= ~VM_IO;
391 WRITE_ONCE(vma->vm_flags, newflags);
392 }
393}
394
395/*
396 * mlock_fixup - handle mlock[all]/munlock[all] requests.
397 *
398 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
399 * munlock is a no-op. However, for some special vmas, we go ahead and
400 * populate the ptes.
401 *
402 * For vmas that pass the filters, merge/split as appropriate.
403 */
404static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
405 unsigned long start, unsigned long end, vm_flags_t newflags)
406{
407 struct mm_struct *mm = vma->vm_mm;
408 pgoff_t pgoff;
409 int nr_pages;
410 int ret = 0;
411 vm_flags_t oldflags = vma->vm_flags;
412
413 if (newflags == oldflags || (oldflags & VM_SPECIAL) ||
414 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm) ||
415 vma_is_dax(vma) || vma_is_secretmem(vma))
416 /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
417 goto out;
418
419 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
420 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
421 vma->vm_file, pgoff, vma_policy(vma),
422 vma->vm_userfaultfd_ctx, anon_vma_name(vma));
423 if (*prev) {
424 vma = *prev;
425 goto success;
426 }
427
428 if (start != vma->vm_start) {
429 ret = split_vma(mm, vma, start, 1);
430 if (ret)
431 goto out;
432 }
433
434 if (end != vma->vm_end) {
435 ret = split_vma(mm, vma, end, 0);
436 if (ret)
437 goto out;
438 }
439
440success:
441 /*
442 * Keep track of amount of locked VM.
443 */
444 nr_pages = (end - start) >> PAGE_SHIFT;
445 if (!(newflags & VM_LOCKED))
446 nr_pages = -nr_pages;
447 else if (oldflags & VM_LOCKED)
448 nr_pages = 0;
449 mm->locked_vm += nr_pages;
450
451 /*
452 * vm_flags is protected by the mmap_lock held in write mode.
453 * It's okay if try_to_unmap_one unmaps a page just after we
454 * set VM_LOCKED, populate_vma_page_range will bring it back.
455 */
456
457 if ((newflags & VM_LOCKED) && (oldflags & VM_LOCKED)) {
458 /* No work to do, and mlocking twice would be wrong */
459 vma->vm_flags = newflags;
460 } else {
461 mlock_vma_pages_range(vma, start, end, newflags);
462 }
463out:
464 *prev = vma;
465 return ret;
466}
467
468static int apply_vma_lock_flags(unsigned long start, size_t len,
469 vm_flags_t flags)
470{
471 unsigned long nstart, end, tmp;
472 struct vm_area_struct *vma, *prev;
473 int error;
474 MA_STATE(mas, ¤t->mm->mm_mt, start, start);
475
476 VM_BUG_ON(offset_in_page(start));
477 VM_BUG_ON(len != PAGE_ALIGN(len));
478 end = start + len;
479 if (end < start)
480 return -EINVAL;
481 if (end == start)
482 return 0;
483 vma = mas_walk(&mas);
484 if (!vma)
485 return -ENOMEM;
486
487 if (start > vma->vm_start)
488 prev = vma;
489 else
490 prev = mas_prev(&mas, 0);
491
492 for (nstart = start ; ; ) {
493 vm_flags_t newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
494
495 newflags |= flags;
496
497 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
498 tmp = vma->vm_end;
499 if (tmp > end)
500 tmp = end;
501 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
502 if (error)
503 break;
504 nstart = tmp;
505 if (nstart < prev->vm_end)
506 nstart = prev->vm_end;
507 if (nstart >= end)
508 break;
509
510 vma = find_vma(prev->vm_mm, prev->vm_end);
511 if (!vma || vma->vm_start != nstart) {
512 error = -ENOMEM;
513 break;
514 }
515 }
516 return error;
517}
518
519/*
520 * Go through vma areas and sum size of mlocked
521 * vma pages, as return value.
522 * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT)
523 * is also counted.
524 * Return value: previously mlocked page counts
525 */
526static unsigned long count_mm_mlocked_page_nr(struct mm_struct *mm,
527 unsigned long start, size_t len)
528{
529 struct vm_area_struct *vma;
530 unsigned long count = 0;
531 unsigned long end;
532 VMA_ITERATOR(vmi, mm, start);
533
534 /* Don't overflow past ULONG_MAX */
535 if (unlikely(ULONG_MAX - len < start))
536 end = ULONG_MAX;
537 else
538 end = start + len;
539
540 for_each_vma_range(vmi, vma, end) {
541 if (vma->vm_flags & VM_LOCKED) {
542 if (start > vma->vm_start)
543 count -= (start - vma->vm_start);
544 if (end < vma->vm_end) {
545 count += end - vma->vm_start;
546 break;
547 }
548 count += vma->vm_end - vma->vm_start;
549 }
550 }
551
552 return count >> PAGE_SHIFT;
553}
554
555/*
556 * convert get_user_pages() return value to posix mlock() error
557 */
558static int __mlock_posix_error_return(long retval)
559{
560 if (retval == -EFAULT)
561 retval = -ENOMEM;
562 else if (retval == -ENOMEM)
563 retval = -EAGAIN;
564 return retval;
565}
566
567static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
568{
569 unsigned long locked;
570 unsigned long lock_limit;
571 int error = -ENOMEM;
572
573 start = untagged_addr(start);
574
575 if (!can_do_mlock())
576 return -EPERM;
577
578 len = PAGE_ALIGN(len + (offset_in_page(start)));
579 start &= PAGE_MASK;
580
581 lock_limit = rlimit(RLIMIT_MEMLOCK);
582 lock_limit >>= PAGE_SHIFT;
583 locked = len >> PAGE_SHIFT;
584
585 if (mmap_write_lock_killable(current->mm))
586 return -EINTR;
587
588 locked += current->mm->locked_vm;
589 if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) {
590 /*
591 * It is possible that the regions requested intersect with
592 * previously mlocked areas, that part area in "mm->locked_vm"
593 * should not be counted to new mlock increment count. So check
594 * and adjust locked count if necessary.
595 */
596 locked -= count_mm_mlocked_page_nr(current->mm,
597 start, len);
598 }
599
600 /* check against resource limits */
601 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
602 error = apply_vma_lock_flags(start, len, flags);
603
604 mmap_write_unlock(current->mm);
605 if (error)
606 return error;
607
608 error = __mm_populate(start, len, 0);
609 if (error)
610 return __mlock_posix_error_return(error);
611 return 0;
612}
613
614SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
615{
616 return do_mlock(start, len, VM_LOCKED);
617}
618
619SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
620{
621 vm_flags_t vm_flags = VM_LOCKED;
622
623 if (flags & ~MLOCK_ONFAULT)
624 return -EINVAL;
625
626 if (flags & MLOCK_ONFAULT)
627 vm_flags |= VM_LOCKONFAULT;
628
629 return do_mlock(start, len, vm_flags);
630}
631
632SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
633{
634 int ret;
635
636 start = untagged_addr(start);
637
638 len = PAGE_ALIGN(len + (offset_in_page(start)));
639 start &= PAGE_MASK;
640
641 if (mmap_write_lock_killable(current->mm))
642 return -EINTR;
643 ret = apply_vma_lock_flags(start, len, 0);
644 mmap_write_unlock(current->mm);
645
646 return ret;
647}
648
649/*
650 * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
651 * and translate into the appropriate modifications to mm->def_flags and/or the
652 * flags for all current VMAs.
653 *
654 * There are a couple of subtleties with this. If mlockall() is called multiple
655 * times with different flags, the values do not necessarily stack. If mlockall
656 * is called once including the MCL_FUTURE flag and then a second time without
657 * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
658 */
659static int apply_mlockall_flags(int flags)
660{
661 MA_STATE(mas, ¤t->mm->mm_mt, 0, 0);
662 struct vm_area_struct *vma, *prev = NULL;
663 vm_flags_t to_add = 0;
664
665 current->mm->def_flags &= VM_LOCKED_CLEAR_MASK;
666 if (flags & MCL_FUTURE) {
667 current->mm->def_flags |= VM_LOCKED;
668
669 if (flags & MCL_ONFAULT)
670 current->mm->def_flags |= VM_LOCKONFAULT;
671
672 if (!(flags & MCL_CURRENT))
673 goto out;
674 }
675
676 if (flags & MCL_CURRENT) {
677 to_add |= VM_LOCKED;
678 if (flags & MCL_ONFAULT)
679 to_add |= VM_LOCKONFAULT;
680 }
681
682 mas_for_each(&mas, vma, ULONG_MAX) {
683 vm_flags_t newflags;
684
685 newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
686 newflags |= to_add;
687
688 /* Ignore errors */
689 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
690 mas_pause(&mas);
691 cond_resched();
692 }
693out:
694 return 0;
695}
696
697SYSCALL_DEFINE1(mlockall, int, flags)
698{
699 unsigned long lock_limit;
700 int ret;
701
702 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)) ||
703 flags == MCL_ONFAULT)
704 return -EINVAL;
705
706 if (!can_do_mlock())
707 return -EPERM;
708
709 lock_limit = rlimit(RLIMIT_MEMLOCK);
710 lock_limit >>= PAGE_SHIFT;
711
712 if (mmap_write_lock_killable(current->mm))
713 return -EINTR;
714
715 ret = -ENOMEM;
716 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
717 capable(CAP_IPC_LOCK))
718 ret = apply_mlockall_flags(flags);
719 mmap_write_unlock(current->mm);
720 if (!ret && (flags & MCL_CURRENT))
721 mm_populate(0, TASK_SIZE);
722
723 return ret;
724}
725
726SYSCALL_DEFINE0(munlockall)
727{
728 int ret;
729
730 if (mmap_write_lock_killable(current->mm))
731 return -EINTR;
732 ret = apply_mlockall_flags(0);
733 mmap_write_unlock(current->mm);
734 return ret;
735}
736
737/*
738 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
739 * shm segments) get accounted against the user_struct instead.
740 */
741static DEFINE_SPINLOCK(shmlock_user_lock);
742
743int user_shm_lock(size_t size, struct ucounts *ucounts)
744{
745 unsigned long lock_limit, locked;
746 long memlock;
747 int allowed = 0;
748
749 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
750 lock_limit = rlimit(RLIMIT_MEMLOCK);
751 if (lock_limit != RLIM_INFINITY)
752 lock_limit >>= PAGE_SHIFT;
753 spin_lock(&shmlock_user_lock);
754 memlock = inc_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
755
756 if ((memlock == LONG_MAX || memlock > lock_limit) && !capable(CAP_IPC_LOCK)) {
757 dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
758 goto out;
759 }
760 if (!get_ucounts(ucounts)) {
761 dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
762 allowed = 0;
763 goto out;
764 }
765 allowed = 1;
766out:
767 spin_unlock(&shmlock_user_lock);
768 return allowed;
769}
770
771void user_shm_unlock(size_t size, struct ucounts *ucounts)
772{
773 spin_lock(&shmlock_user_lock);
774 dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, (size + PAGE_SIZE - 1) >> PAGE_SHIFT);
775 spin_unlock(&shmlock_user_lock);
776 put_ucounts(ucounts);
777}