Line data Source code
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 :
31 : struct mlock_fbatch {
32 : local_lock_t lock;
33 : struct folio_batch fbatch;
34 : };
35 :
36 : static DEFINE_PER_CPU(struct mlock_fbatch, mlock_fbatch) = {
37 : .lock = INIT_LOCAL_LOCK(lock),
38 : };
39 :
40 0 : bool can_do_mlock(void)
41 : {
42 0 : if (rlimit(RLIMIT_MEMLOCK) != 0)
43 : return true;
44 0 : if (capable(CAP_IPC_LOCK))
45 : return true;
46 0 : return false;
47 : }
48 : EXPORT_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 :
61 0 : static 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 0 : if (!folio_test_clear_lru(folio))
65 : return lruvec;
66 :
67 0 : lruvec = folio_lruvec_relock_irq(folio, lruvec);
68 :
69 0 : 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 0 : if (folio_test_unevictable(folio)) {
76 0 : lruvec_del_folio(lruvec, folio);
77 0 : folio_clear_unevictable(folio);
78 0 : lruvec_add_folio(lruvec, folio);
79 :
80 0 : __count_vm_events(UNEVICTABLE_PGRESCUED,
81 : folio_nr_pages(folio));
82 : }
83 : goto out;
84 : }
85 :
86 0 : if (folio_test_unevictable(folio)) {
87 0 : if (folio_test_mlocked(folio))
88 0 : folio->mlock_count++;
89 : goto out;
90 : }
91 :
92 0 : lruvec_del_folio(lruvec, folio);
93 0 : folio_clear_active(folio);
94 0 : folio_set_unevictable(folio);
95 0 : folio->mlock_count = !!folio_test_mlocked(folio);
96 0 : lruvec_add_folio(lruvec, folio);
97 0 : __count_vm_events(UNEVICTABLE_PGCULLED, folio_nr_pages(folio));
98 : out:
99 0 : folio_set_lru(folio);
100 0 : return lruvec;
101 : }
102 :
103 0 : static struct lruvec *__mlock_new_folio(struct folio *folio, struct lruvec *lruvec)
104 : {
105 : VM_BUG_ON_FOLIO(folio_test_lru(folio), folio);
106 :
107 0 : lruvec = folio_lruvec_relock_irq(folio, lruvec);
108 :
109 : /* As above, this is a little surprising, but possible */
110 0 : if (unlikely(folio_evictable(folio)))
111 : goto out;
112 :
113 0 : folio_set_unevictable(folio);
114 0 : folio->mlock_count = !!folio_test_mlocked(folio);
115 0 : __count_vm_events(UNEVICTABLE_PGCULLED, folio_nr_pages(folio));
116 : out:
117 0 : lruvec_add_folio(lruvec, folio);
118 0 : folio_set_lru(folio);
119 0 : return lruvec;
120 : }
121 :
122 0 : static struct lruvec *__munlock_folio(struct folio *folio, struct lruvec *lruvec)
123 : {
124 0 : int nr_pages = folio_nr_pages(folio);
125 0 : bool isolated = false;
126 :
127 0 : if (!folio_test_clear_lru(folio))
128 : goto munlock;
129 :
130 0 : isolated = true;
131 0 : lruvec = folio_lruvec_relock_irq(folio, lruvec);
132 :
133 0 : if (folio_test_unevictable(folio)) {
134 : /* Then mlock_count is maintained, but might undercount */
135 0 : if (folio->mlock_count)
136 0 : folio->mlock_count--;
137 0 : if (folio->mlock_count)
138 : goto out;
139 : }
140 : /* else assume that was the last mlock: reclaim will fix it if not */
141 :
142 : munlock:
143 0 : if (folio_test_clear_mlocked(folio)) {
144 0 : __zone_stat_mod_folio(folio, NR_MLOCK, -nr_pages);
145 0 : if (isolated || !folio_test_unevictable(folio))
146 0 : __count_vm_events(UNEVICTABLE_PGMUNLOCKED, nr_pages);
147 : else
148 0 : __count_vm_events(UNEVICTABLE_PGSTRANDED, nr_pages);
149 : }
150 :
151 : /* folio_evictable() has to be checked *after* clearing Mlocked */
152 0 : if (isolated && folio_test_unevictable(folio) && folio_evictable(folio)) {
153 0 : lruvec_del_folio(lruvec, folio);
154 0 : folio_clear_unevictable(folio);
155 0 : lruvec_add_folio(lruvec, folio);
156 0 : __count_vm_events(UNEVICTABLE_PGRESCUED, nr_pages);
157 : }
158 : out:
159 0 : if (isolated)
160 : folio_set_lru(folio);
161 0 : 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
169 : static inline struct folio *mlock_lru(struct folio *folio)
170 : {
171 0 : return (struct folio *)((unsigned long)folio + LRU_FOLIO);
172 : }
173 :
174 : static inline struct folio *mlock_new(struct folio *folio)
175 : {
176 0 : 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 : */
186 0 : static void mlock_folio_batch(struct folio_batch *fbatch)
187 : {
188 0 : struct lruvec *lruvec = NULL;
189 : unsigned long mlock;
190 : struct folio *folio;
191 : int i;
192 :
193 0 : for (i = 0; i < folio_batch_count(fbatch); i++) {
194 0 : folio = fbatch->folios[i];
195 0 : mlock = (unsigned long)folio & (LRU_FOLIO | NEW_FOLIO);
196 0 : folio = (struct folio *)((unsigned long)folio - mlock);
197 0 : fbatch->folios[i] = folio;
198 :
199 0 : if (mlock & LRU_FOLIO)
200 0 : lruvec = __mlock_folio(folio, lruvec);
201 0 : else if (mlock & NEW_FOLIO)
202 0 : lruvec = __mlock_new_folio(folio, lruvec);
203 : else
204 0 : lruvec = __munlock_folio(folio, lruvec);
205 : }
206 :
207 0 : if (lruvec)
208 0 : unlock_page_lruvec_irq(lruvec);
209 0 : folios_put(fbatch->folios, folio_batch_count(fbatch));
210 0 : folio_batch_reinit(fbatch);
211 0 : }
212 :
213 0 : void mlock_drain_local(void)
214 : {
215 : struct folio_batch *fbatch;
216 :
217 0 : local_lock(&mlock_fbatch.lock);
218 0 : fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
219 0 : if (folio_batch_count(fbatch))
220 0 : mlock_folio_batch(fbatch);
221 0 : local_unlock(&mlock_fbatch.lock);
222 0 : }
223 :
224 0 : void mlock_drain_remote(int cpu)
225 : {
226 : struct folio_batch *fbatch;
227 :
228 0 : WARN_ON_ONCE(cpu_online(cpu));
229 0 : fbatch = &per_cpu(mlock_fbatch.fbatch, cpu);
230 0 : if (folio_batch_count(fbatch))
231 0 : mlock_folio_batch(fbatch);
232 0 : }
233 :
234 0 : bool need_mlock_drain(int cpu)
235 : {
236 0 : return folio_batch_count(&per_cpu(mlock_fbatch.fbatch, cpu));
237 : }
238 :
239 : /**
240 : * mlock_folio - mlock a folio already on (or temporarily off) LRU
241 : * @folio: folio to be mlocked.
242 : */
243 0 : void mlock_folio(struct folio *folio)
244 : {
245 : struct folio_batch *fbatch;
246 :
247 0 : local_lock(&mlock_fbatch.lock);
248 0 : fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
249 :
250 0 : if (!folio_test_set_mlocked(folio)) {
251 0 : int nr_pages = folio_nr_pages(folio);
252 :
253 0 : zone_stat_mod_folio(folio, NR_MLOCK, nr_pages);
254 0 : __count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
255 : }
256 :
257 0 : folio_get(folio);
258 0 : if (!folio_batch_add(fbatch, mlock_lru(folio)) ||
259 0 : folio_test_large(folio) || lru_cache_disabled())
260 0 : mlock_folio_batch(fbatch);
261 0 : local_unlock(&mlock_fbatch.lock);
262 0 : }
263 :
264 : /**
265 : * mlock_new_folio - mlock a newly allocated folio not yet on LRU
266 : * @folio: folio to be mlocked, either normal or a THP head.
267 : */
268 0 : void mlock_new_folio(struct folio *folio)
269 : {
270 : struct folio_batch *fbatch;
271 0 : int nr_pages = folio_nr_pages(folio);
272 :
273 0 : local_lock(&mlock_fbatch.lock);
274 0 : fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
275 0 : folio_set_mlocked(folio);
276 :
277 0 : zone_stat_mod_folio(folio, NR_MLOCK, nr_pages);
278 0 : __count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
279 :
280 0 : folio_get(folio);
281 0 : if (!folio_batch_add(fbatch, mlock_new(folio)) ||
282 0 : folio_test_large(folio) || lru_cache_disabled())
283 0 : mlock_folio_batch(fbatch);
284 0 : local_unlock(&mlock_fbatch.lock);
285 0 : }
286 :
287 : /**
288 : * munlock_folio - munlock a folio
289 : * @folio: folio to be munlocked, either normal or a THP head.
290 : */
291 0 : void munlock_folio(struct folio *folio)
292 : {
293 : struct folio_batch *fbatch;
294 :
295 0 : local_lock(&mlock_fbatch.lock);
296 0 : fbatch = this_cpu_ptr(&mlock_fbatch.fbatch);
297 : /*
298 : * folio_test_clear_mlocked(folio) must be left to __munlock_folio(),
299 : * which will check whether the folio is multiply mlocked.
300 : */
301 0 : folio_get(folio);
302 0 : if (!folio_batch_add(fbatch, folio) ||
303 0 : folio_test_large(folio) || lru_cache_disabled())
304 0 : mlock_folio_batch(fbatch);
305 0 : local_unlock(&mlock_fbatch.lock);
306 0 : }
307 :
308 0 : static int mlock_pte_range(pmd_t *pmd, unsigned long addr,
309 : unsigned long end, struct mm_walk *walk)
310 :
311 : {
312 0 : struct vm_area_struct *vma = walk->vma;
313 : spinlock_t *ptl;
314 : pte_t *start_pte, *pte;
315 : pte_t ptent;
316 : struct folio *folio;
317 :
318 0 : ptl = pmd_trans_huge_lock(pmd, vma);
319 : if (ptl) {
320 : if (!pmd_present(*pmd))
321 : goto out;
322 : if (is_huge_zero_pmd(*pmd))
323 : goto out;
324 : folio = page_folio(pmd_page(*pmd));
325 : if (vma->vm_flags & VM_LOCKED)
326 : mlock_folio(folio);
327 : else
328 : munlock_folio(folio);
329 : goto out;
330 : }
331 :
332 0 : start_pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
333 0 : if (!start_pte) {
334 0 : walk->action = ACTION_AGAIN;
335 0 : return 0;
336 : }
337 0 : for (pte = start_pte; addr != end; pte++, addr += PAGE_SIZE) {
338 0 : ptent = ptep_get(pte);
339 0 : if (!pte_present(ptent))
340 0 : continue;
341 0 : folio = vm_normal_folio(vma, addr, ptent);
342 0 : if (!folio || folio_is_zone_device(folio))
343 0 : continue;
344 0 : if (folio_test_large(folio))
345 0 : continue;
346 0 : if (vma->vm_flags & VM_LOCKED)
347 0 : mlock_folio(folio);
348 : else
349 0 : munlock_folio(folio);
350 : }
351 : pte_unmap(start_pte);
352 : out:
353 0 : spin_unlock(ptl);
354 0 : cond_resched();
355 0 : return 0;
356 : }
357 :
358 : /*
359 : * mlock_vma_pages_range() - mlock any pages already in the range,
360 : * or munlock all pages in the range.
361 : * @vma - vma containing range to be mlock()ed or munlock()ed
362 : * @start - start address in @vma of the range
363 : * @end - end of range in @vma
364 : * @newflags - the new set of flags for @vma.
365 : *
366 : * Called for mlock(), mlock2() and mlockall(), to set @vma VM_LOCKED;
367 : * called for munlock() and munlockall(), to clear VM_LOCKED from @vma.
368 : */
369 0 : static void mlock_vma_pages_range(struct vm_area_struct *vma,
370 : unsigned long start, unsigned long end, vm_flags_t newflags)
371 : {
372 : static const struct mm_walk_ops mlock_walk_ops = {
373 : .pmd_entry = mlock_pte_range,
374 : };
375 :
376 : /*
377 : * There is a slight chance that concurrent page migration,
378 : * or page reclaim finding a page of this now-VM_LOCKED vma,
379 : * will call mlock_vma_folio() and raise page's mlock_count:
380 : * double counting, leaving the page unevictable indefinitely.
381 : * Communicate this danger to mlock_vma_folio() with VM_IO,
382 : * which is a VM_SPECIAL flag not allowed on VM_LOCKED vmas.
383 : * mmap_lock is held in write mode here, so this weird
384 : * combination should not be visible to other mmap_lock users;
385 : * but WRITE_ONCE so rmap walkers must see VM_IO if VM_LOCKED.
386 : */
387 0 : if (newflags & VM_LOCKED)
388 0 : newflags |= VM_IO;
389 0 : vm_flags_reset_once(vma, newflags);
390 :
391 0 : lru_add_drain();
392 0 : walk_page_range(vma->vm_mm, start, end, &mlock_walk_ops, NULL);
393 0 : lru_add_drain();
394 :
395 0 : if (newflags & VM_IO) {
396 0 : newflags &= ~VM_IO;
397 : vm_flags_reset_once(vma, newflags);
398 : }
399 0 : }
400 :
401 : /*
402 : * mlock_fixup - handle mlock[all]/munlock[all] requests.
403 : *
404 : * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
405 : * munlock is a no-op. However, for some special vmas, we go ahead and
406 : * populate the ptes.
407 : *
408 : * For vmas that pass the filters, merge/split as appropriate.
409 : */
410 0 : static int mlock_fixup(struct vma_iterator *vmi, struct vm_area_struct *vma,
411 : struct vm_area_struct **prev, unsigned long start,
412 : unsigned long end, vm_flags_t newflags)
413 : {
414 0 : struct mm_struct *mm = vma->vm_mm;
415 : pgoff_t pgoff;
416 : int nr_pages;
417 0 : int ret = 0;
418 0 : vm_flags_t oldflags = vma->vm_flags;
419 :
420 0 : if (newflags == oldflags || (oldflags & VM_SPECIAL) ||
421 0 : is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm) ||
422 0 : vma_is_dax(vma) || vma_is_secretmem(vma))
423 : /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
424 : goto out;
425 :
426 0 : pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
427 0 : *prev = vma_merge(vmi, mm, *prev, start, end, newflags,
428 : vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
429 : vma->vm_userfaultfd_ctx, anon_vma_name(vma));
430 0 : if (*prev) {
431 : vma = *prev;
432 : goto success;
433 : }
434 :
435 0 : if (start != vma->vm_start) {
436 0 : ret = split_vma(vmi, vma, start, 1);
437 0 : if (ret)
438 : goto out;
439 : }
440 :
441 0 : if (end != vma->vm_end) {
442 0 : ret = split_vma(vmi, vma, end, 0);
443 0 : if (ret)
444 : goto out;
445 : }
446 :
447 : success:
448 : /*
449 : * Keep track of amount of locked VM.
450 : */
451 0 : nr_pages = (end - start) >> PAGE_SHIFT;
452 0 : if (!(newflags & VM_LOCKED))
453 0 : nr_pages = -nr_pages;
454 0 : else if (oldflags & VM_LOCKED)
455 0 : nr_pages = 0;
456 0 : mm->locked_vm += nr_pages;
457 :
458 : /*
459 : * vm_flags is protected by the mmap_lock held in write mode.
460 : * It's okay if try_to_unmap_one unmaps a page just after we
461 : * set VM_LOCKED, populate_vma_page_range will bring it back.
462 : */
463 :
464 0 : if ((newflags & VM_LOCKED) && (oldflags & VM_LOCKED)) {
465 : /* No work to do, and mlocking twice would be wrong */
466 : vm_flags_reset(vma, newflags);
467 : } else {
468 0 : mlock_vma_pages_range(vma, start, end, newflags);
469 : }
470 : out:
471 0 : *prev = vma;
472 0 : return ret;
473 : }
474 :
475 0 : static int apply_vma_lock_flags(unsigned long start, size_t len,
476 : vm_flags_t flags)
477 : {
478 : unsigned long nstart, end, tmp;
479 : struct vm_area_struct *vma, *prev;
480 : int error;
481 0 : VMA_ITERATOR(vmi, current->mm, start);
482 :
483 : VM_BUG_ON(offset_in_page(start));
484 : VM_BUG_ON(len != PAGE_ALIGN(len));
485 0 : end = start + len;
486 0 : if (end < start)
487 : return -EINVAL;
488 0 : if (end == start)
489 : return 0;
490 0 : vma = vma_iter_load(&vmi);
491 0 : if (!vma)
492 : return -ENOMEM;
493 :
494 0 : prev = vma_prev(&vmi);
495 0 : if (start > vma->vm_start)
496 0 : prev = vma;
497 :
498 0 : nstart = start;
499 0 : tmp = vma->vm_start;
500 0 : for_each_vma_range(vmi, vma, end) {
501 : vm_flags_t newflags;
502 :
503 0 : if (vma->vm_start != tmp)
504 : return -ENOMEM;
505 :
506 0 : newflags = vma->vm_flags & ~VM_LOCKED_MASK;
507 0 : newflags |= flags;
508 : /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
509 0 : tmp = vma->vm_end;
510 0 : if (tmp > end)
511 0 : tmp = end;
512 0 : error = mlock_fixup(&vmi, vma, &prev, nstart, tmp, newflags);
513 0 : if (error)
514 : break;
515 : nstart = tmp;
516 : }
517 :
518 0 : if (vma_iter_end(&vmi) < end)
519 : return -ENOMEM;
520 :
521 0 : return error;
522 : }
523 :
524 : /*
525 : * Go through vma areas and sum size of mlocked
526 : * vma pages, as return value.
527 : * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT)
528 : * is also counted.
529 : * Return value: previously mlocked page counts
530 : */
531 0 : static unsigned long count_mm_mlocked_page_nr(struct mm_struct *mm,
532 : unsigned long start, size_t len)
533 : {
534 : struct vm_area_struct *vma;
535 0 : unsigned long count = 0;
536 : unsigned long end;
537 0 : VMA_ITERATOR(vmi, mm, start);
538 :
539 : /* Don't overflow past ULONG_MAX */
540 0 : if (unlikely(ULONG_MAX - len < start))
541 : end = ULONG_MAX;
542 : else
543 0 : end = start + len;
544 :
545 0 : for_each_vma_range(vmi, vma, end) {
546 0 : if (vma->vm_flags & VM_LOCKED) {
547 0 : if (start > vma->vm_start)
548 0 : count -= (start - vma->vm_start);
549 0 : if (end < vma->vm_end) {
550 0 : count += end - vma->vm_start;
551 0 : break;
552 : }
553 0 : count += vma->vm_end - vma->vm_start;
554 : }
555 : }
556 :
557 0 : return count >> PAGE_SHIFT;
558 : }
559 :
560 : /*
561 : * convert get_user_pages() return value to posix mlock() error
562 : */
563 : static int __mlock_posix_error_return(long retval)
564 : {
565 0 : if (retval == -EFAULT)
566 : retval = -ENOMEM;
567 0 : else if (retval == -ENOMEM)
568 0 : retval = -EAGAIN;
569 0 : return retval;
570 : }
571 :
572 0 : static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
573 : {
574 : unsigned long locked;
575 : unsigned long lock_limit;
576 0 : int error = -ENOMEM;
577 :
578 0 : start = untagged_addr(start);
579 :
580 0 : if (!can_do_mlock())
581 : return -EPERM;
582 :
583 0 : len = PAGE_ALIGN(len + (offset_in_page(start)));
584 0 : start &= PAGE_MASK;
585 :
586 0 : lock_limit = rlimit(RLIMIT_MEMLOCK);
587 0 : lock_limit >>= PAGE_SHIFT;
588 0 : locked = len >> PAGE_SHIFT;
589 :
590 0 : if (mmap_write_lock_killable(current->mm))
591 : return -EINTR;
592 :
593 0 : locked += current->mm->locked_vm;
594 0 : if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) {
595 : /*
596 : * It is possible that the regions requested intersect with
597 : * previously mlocked areas, that part area in "mm->locked_vm"
598 : * should not be counted to new mlock increment count. So check
599 : * and adjust locked count if necessary.
600 : */
601 0 : locked -= count_mm_mlocked_page_nr(current->mm,
602 : start, len);
603 : }
604 :
605 : /* check against resource limits */
606 0 : if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
607 0 : error = apply_vma_lock_flags(start, len, flags);
608 :
609 0 : mmap_write_unlock(current->mm);
610 0 : if (error)
611 : return error;
612 :
613 0 : error = __mm_populate(start, len, 0);
614 0 : if (error)
615 0 : return __mlock_posix_error_return(error);
616 : return 0;
617 : }
618 :
619 0 : SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
620 : {
621 0 : return do_mlock(start, len, VM_LOCKED);
622 : }
623 :
624 0 : SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
625 : {
626 0 : vm_flags_t vm_flags = VM_LOCKED;
627 :
628 0 : if (flags & ~MLOCK_ONFAULT)
629 : return -EINVAL;
630 :
631 0 : if (flags & MLOCK_ONFAULT)
632 0 : vm_flags |= VM_LOCKONFAULT;
633 :
634 0 : return do_mlock(start, len, vm_flags);
635 : }
636 :
637 0 : SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
638 : {
639 : int ret;
640 :
641 0 : start = untagged_addr(start);
642 :
643 0 : len = PAGE_ALIGN(len + (offset_in_page(start)));
644 0 : start &= PAGE_MASK;
645 :
646 0 : if (mmap_write_lock_killable(current->mm))
647 : return -EINTR;
648 0 : ret = apply_vma_lock_flags(start, len, 0);
649 0 : mmap_write_unlock(current->mm);
650 :
651 0 : return ret;
652 : }
653 :
654 : /*
655 : * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
656 : * and translate into the appropriate modifications to mm->def_flags and/or the
657 : * flags for all current VMAs.
658 : *
659 : * There are a couple of subtleties with this. If mlockall() is called multiple
660 : * times with different flags, the values do not necessarily stack. If mlockall
661 : * is called once including the MCL_FUTURE flag and then a second time without
662 : * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
663 : */
664 0 : static int apply_mlockall_flags(int flags)
665 : {
666 0 : VMA_ITERATOR(vmi, current->mm, 0);
667 0 : struct vm_area_struct *vma, *prev = NULL;
668 0 : vm_flags_t to_add = 0;
669 :
670 0 : current->mm->def_flags &= ~VM_LOCKED_MASK;
671 0 : if (flags & MCL_FUTURE) {
672 0 : current->mm->def_flags |= VM_LOCKED;
673 :
674 0 : if (flags & MCL_ONFAULT)
675 0 : current->mm->def_flags |= VM_LOCKONFAULT;
676 :
677 0 : if (!(flags & MCL_CURRENT))
678 : goto out;
679 : }
680 :
681 0 : if (flags & MCL_CURRENT) {
682 0 : to_add |= VM_LOCKED;
683 0 : if (flags & MCL_ONFAULT)
684 0 : to_add |= VM_LOCKONFAULT;
685 : }
686 :
687 0 : for_each_vma(vmi, vma) {
688 : vm_flags_t newflags;
689 :
690 0 : newflags = vma->vm_flags & ~VM_LOCKED_MASK;
691 0 : newflags |= to_add;
692 :
693 : /* Ignore errors */
694 0 : mlock_fixup(&vmi, vma, &prev, vma->vm_start, vma->vm_end,
695 : newflags);
696 0 : cond_resched();
697 : }
698 : out:
699 0 : return 0;
700 : }
701 :
702 0 : SYSCALL_DEFINE1(mlockall, int, flags)
703 : {
704 : unsigned long lock_limit;
705 : int ret;
706 :
707 0 : if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)) ||
708 : flags == MCL_ONFAULT)
709 : return -EINVAL;
710 :
711 0 : if (!can_do_mlock())
712 : return -EPERM;
713 :
714 0 : lock_limit = rlimit(RLIMIT_MEMLOCK);
715 0 : lock_limit >>= PAGE_SHIFT;
716 :
717 0 : if (mmap_write_lock_killable(current->mm))
718 : return -EINTR;
719 :
720 0 : ret = -ENOMEM;
721 0 : if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
722 0 : capable(CAP_IPC_LOCK))
723 0 : ret = apply_mlockall_flags(flags);
724 0 : mmap_write_unlock(current->mm);
725 0 : if (!ret && (flags & MCL_CURRENT))
726 0 : mm_populate(0, TASK_SIZE);
727 :
728 0 : return ret;
729 : }
730 :
731 0 : SYSCALL_DEFINE0(munlockall)
732 : {
733 : int ret;
734 :
735 0 : if (mmap_write_lock_killable(current->mm))
736 : return -EINTR;
737 0 : ret = apply_mlockall_flags(0);
738 0 : mmap_write_unlock(current->mm);
739 0 : return ret;
740 : }
741 :
742 : /*
743 : * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
744 : * shm segments) get accounted against the user_struct instead.
745 : */
746 : static DEFINE_SPINLOCK(shmlock_user_lock);
747 :
748 0 : int user_shm_lock(size_t size, struct ucounts *ucounts)
749 : {
750 : unsigned long lock_limit, locked;
751 : long memlock;
752 0 : int allowed = 0;
753 :
754 0 : locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
755 0 : lock_limit = rlimit(RLIMIT_MEMLOCK);
756 0 : if (lock_limit != RLIM_INFINITY)
757 0 : lock_limit >>= PAGE_SHIFT;
758 0 : spin_lock(&shmlock_user_lock);
759 0 : memlock = inc_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
760 :
761 0 : if ((memlock == LONG_MAX || memlock > lock_limit) && !capable(CAP_IPC_LOCK)) {
762 0 : dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
763 0 : goto out;
764 : }
765 0 : if (!get_ucounts(ucounts)) {
766 0 : dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
767 0 : allowed = 0;
768 0 : goto out;
769 : }
770 : allowed = 1;
771 : out:
772 0 : spin_unlock(&shmlock_user_lock);
773 0 : return allowed;
774 : }
775 :
776 0 : void user_shm_unlock(size_t size, struct ucounts *ucounts)
777 : {
778 0 : spin_lock(&shmlock_user_lock);
779 0 : dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, (size + PAGE_SIZE - 1) >> PAGE_SHIFT);
780 0 : spin_unlock(&shmlock_user_lock);
781 0 : put_ucounts(ucounts);
782 0 : }
|
