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 : struct folio *folio;
316 :
317 0 : 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 : folio = page_folio(pmd_page(*pmd));
324 : if (vma->vm_flags & VM_LOCKED)
325 : mlock_folio(folio);
326 : else
327 : munlock_folio(folio);
328 : goto out;
329 : }
330 :
331 0 : start_pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
332 0 : for (pte = start_pte; addr != end; pte++, addr += PAGE_SIZE) {
333 0 : if (!pte_present(*pte))
334 0 : continue;
335 0 : folio = vm_normal_folio(vma, addr, *pte);
336 0 : if (!folio || folio_is_zone_device(folio))
337 0 : continue;
338 0 : if (folio_test_large(folio))
339 0 : continue;
340 0 : if (vma->vm_flags & VM_LOCKED)
341 0 : mlock_folio(folio);
342 : else
343 0 : munlock_folio(folio);
344 : }
345 : pte_unmap(start_pte);
346 : out:
347 0 : spin_unlock(ptl);
348 0 : cond_resched();
349 0 : 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 : */
363 0 : static 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_folio() and raise page's mlock_count:
374 : * double counting, leaving the page unevictable indefinitely.
375 : * Communicate this danger to mlock_vma_folio() 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 0 : if (newflags & VM_LOCKED)
382 0 : newflags |= VM_IO;
383 0 : vm_flags_reset_once(vma, newflags);
384 :
385 0 : lru_add_drain();
386 0 : walk_page_range(vma->vm_mm, start, end, &mlock_walk_ops, NULL);
387 0 : lru_add_drain();
388 :
389 0 : if (newflags & VM_IO) {
390 0 : newflags &= ~VM_IO;
391 : vm_flags_reset_once(vma, newflags);
392 : }
393 0 : }
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 : */
404 0 : static int mlock_fixup(struct vma_iterator *vmi, struct vm_area_struct *vma,
405 : struct vm_area_struct **prev, unsigned long start,
406 : unsigned long end, vm_flags_t newflags)
407 : {
408 0 : struct mm_struct *mm = vma->vm_mm;
409 : pgoff_t pgoff;
410 : int nr_pages;
411 0 : int ret = 0;
412 0 : vm_flags_t oldflags = vma->vm_flags;
413 :
414 0 : if (newflags == oldflags || (oldflags & VM_SPECIAL) ||
415 0 : is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm) ||
416 0 : vma_is_dax(vma) || vma_is_secretmem(vma))
417 : /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
418 : goto out;
419 :
420 0 : pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
421 0 : *prev = vma_merge(vmi, mm, *prev, start, end, newflags,
422 : vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
423 : vma->vm_userfaultfd_ctx, anon_vma_name(vma));
424 0 : if (*prev) {
425 : vma = *prev;
426 : goto success;
427 : }
428 :
429 0 : if (start != vma->vm_start) {
430 0 : ret = split_vma(vmi, vma, start, 1);
431 0 : if (ret)
432 : goto out;
433 : }
434 :
435 0 : if (end != vma->vm_end) {
436 0 : ret = split_vma(vmi, vma, end, 0);
437 0 : if (ret)
438 : goto out;
439 : }
440 :
441 : success:
442 : /*
443 : * Keep track of amount of locked VM.
444 : */
445 0 : nr_pages = (end - start) >> PAGE_SHIFT;
446 0 : if (!(newflags & VM_LOCKED))
447 0 : nr_pages = -nr_pages;
448 0 : else if (oldflags & VM_LOCKED)
449 0 : nr_pages = 0;
450 0 : mm->locked_vm += nr_pages;
451 :
452 : /*
453 : * vm_flags is protected by the mmap_lock held in write mode.
454 : * It's okay if try_to_unmap_one unmaps a page just after we
455 : * set VM_LOCKED, populate_vma_page_range will bring it back.
456 : */
457 :
458 0 : if ((newflags & VM_LOCKED) && (oldflags & VM_LOCKED)) {
459 : /* No work to do, and mlocking twice would be wrong */
460 : vm_flags_reset(vma, newflags);
461 : } else {
462 0 : mlock_vma_pages_range(vma, start, end, newflags);
463 : }
464 : out:
465 0 : *prev = vma;
466 0 : return ret;
467 : }
468 :
469 0 : static int apply_vma_lock_flags(unsigned long start, size_t len,
470 : vm_flags_t flags)
471 : {
472 : unsigned long nstart, end, tmp;
473 : struct vm_area_struct *vma, *prev;
474 : int error;
475 0 : VMA_ITERATOR(vmi, current->mm, start);
476 :
477 : VM_BUG_ON(offset_in_page(start));
478 : VM_BUG_ON(len != PAGE_ALIGN(len));
479 0 : end = start + len;
480 0 : if (end < start)
481 : return -EINVAL;
482 0 : if (end == start)
483 : return 0;
484 0 : vma = vma_iter_load(&vmi);
485 0 : if (!vma)
486 : return -ENOMEM;
487 :
488 0 : prev = vma_prev(&vmi);
489 0 : if (start > vma->vm_start)
490 0 : prev = vma;
491 :
492 0 : nstart = start;
493 0 : tmp = vma->vm_start;
494 0 : for_each_vma_range(vmi, vma, end) {
495 : vm_flags_t newflags;
496 :
497 0 : if (vma->vm_start != tmp)
498 : return -ENOMEM;
499 :
500 0 : newflags = vma->vm_flags & ~VM_LOCKED_MASK;
501 0 : newflags |= flags;
502 : /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
503 0 : tmp = vma->vm_end;
504 0 : if (tmp > end)
505 0 : tmp = end;
506 0 : error = mlock_fixup(&vmi, vma, &prev, nstart, tmp, newflags);
507 0 : if (error)
508 : break;
509 : nstart = tmp;
510 : }
511 :
512 0 : if (vma_iter_end(&vmi) < end)
513 : return -ENOMEM;
514 :
515 0 : return error;
516 : }
517 :
518 : /*
519 : * Go through vma areas and sum size of mlocked
520 : * vma pages, as return value.
521 : * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT)
522 : * is also counted.
523 : * Return value: previously mlocked page counts
524 : */
525 0 : static unsigned long count_mm_mlocked_page_nr(struct mm_struct *mm,
526 : unsigned long start, size_t len)
527 : {
528 : struct vm_area_struct *vma;
529 0 : unsigned long count = 0;
530 : unsigned long end;
531 0 : VMA_ITERATOR(vmi, mm, start);
532 :
533 : /* Don't overflow past ULONG_MAX */
534 0 : if (unlikely(ULONG_MAX - len < start))
535 : end = ULONG_MAX;
536 : else
537 0 : end = start + len;
538 :
539 0 : for_each_vma_range(vmi, vma, end) {
540 0 : if (vma->vm_flags & VM_LOCKED) {
541 0 : if (start > vma->vm_start)
542 0 : count -= (start - vma->vm_start);
543 0 : if (end < vma->vm_end) {
544 0 : count += end - vma->vm_start;
545 0 : break;
546 : }
547 0 : count += vma->vm_end - vma->vm_start;
548 : }
549 : }
550 :
551 0 : return count >> PAGE_SHIFT;
552 : }
553 :
554 : /*
555 : * convert get_user_pages() return value to posix mlock() error
556 : */
557 : static int __mlock_posix_error_return(long retval)
558 : {
559 0 : if (retval == -EFAULT)
560 : retval = -ENOMEM;
561 0 : else if (retval == -ENOMEM)
562 0 : retval = -EAGAIN;
563 0 : return retval;
564 : }
565 :
566 0 : static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
567 : {
568 : unsigned long locked;
569 : unsigned long lock_limit;
570 0 : int error = -ENOMEM;
571 :
572 0 : start = untagged_addr(start);
573 :
574 0 : if (!can_do_mlock())
575 : return -EPERM;
576 :
577 0 : len = PAGE_ALIGN(len + (offset_in_page(start)));
578 0 : start &= PAGE_MASK;
579 :
580 0 : lock_limit = rlimit(RLIMIT_MEMLOCK);
581 0 : lock_limit >>= PAGE_SHIFT;
582 0 : locked = len >> PAGE_SHIFT;
583 :
584 0 : if (mmap_write_lock_killable(current->mm))
585 : return -EINTR;
586 :
587 0 : locked += current->mm->locked_vm;
588 0 : if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) {
589 : /*
590 : * It is possible that the regions requested intersect with
591 : * previously mlocked areas, that part area in "mm->locked_vm"
592 : * should not be counted to new mlock increment count. So check
593 : * and adjust locked count if necessary.
594 : */
595 0 : locked -= count_mm_mlocked_page_nr(current->mm,
596 : start, len);
597 : }
598 :
599 : /* check against resource limits */
600 0 : if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
601 0 : error = apply_vma_lock_flags(start, len, flags);
602 :
603 0 : mmap_write_unlock(current->mm);
604 0 : if (error)
605 : return error;
606 :
607 0 : error = __mm_populate(start, len, 0);
608 0 : if (error)
609 0 : return __mlock_posix_error_return(error);
610 : return 0;
611 : }
612 :
613 0 : SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
614 : {
615 0 : return do_mlock(start, len, VM_LOCKED);
616 : }
617 :
618 0 : SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
619 : {
620 0 : vm_flags_t vm_flags = VM_LOCKED;
621 :
622 0 : if (flags & ~MLOCK_ONFAULT)
623 : return -EINVAL;
624 :
625 0 : if (flags & MLOCK_ONFAULT)
626 0 : vm_flags |= VM_LOCKONFAULT;
627 :
628 0 : return do_mlock(start, len, vm_flags);
629 : }
630 :
631 0 : SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
632 : {
633 : int ret;
634 :
635 0 : start = untagged_addr(start);
636 :
637 0 : len = PAGE_ALIGN(len + (offset_in_page(start)));
638 0 : start &= PAGE_MASK;
639 :
640 0 : if (mmap_write_lock_killable(current->mm))
641 : return -EINTR;
642 0 : ret = apply_vma_lock_flags(start, len, 0);
643 0 : mmap_write_unlock(current->mm);
644 :
645 0 : return ret;
646 : }
647 :
648 : /*
649 : * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
650 : * and translate into the appropriate modifications to mm->def_flags and/or the
651 : * flags for all current VMAs.
652 : *
653 : * There are a couple of subtleties with this. If mlockall() is called multiple
654 : * times with different flags, the values do not necessarily stack. If mlockall
655 : * is called once including the MCL_FUTURE flag and then a second time without
656 : * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
657 : */
658 0 : static int apply_mlockall_flags(int flags)
659 : {
660 0 : VMA_ITERATOR(vmi, current->mm, 0);
661 0 : struct vm_area_struct *vma, *prev = NULL;
662 0 : vm_flags_t to_add = 0;
663 :
664 0 : current->mm->def_flags &= ~VM_LOCKED_MASK;
665 0 : if (flags & MCL_FUTURE) {
666 0 : current->mm->def_flags |= VM_LOCKED;
667 :
668 0 : if (flags & MCL_ONFAULT)
669 0 : current->mm->def_flags |= VM_LOCKONFAULT;
670 :
671 0 : if (!(flags & MCL_CURRENT))
672 : goto out;
673 : }
674 :
675 0 : if (flags & MCL_CURRENT) {
676 0 : to_add |= VM_LOCKED;
677 0 : if (flags & MCL_ONFAULT)
678 0 : to_add |= VM_LOCKONFAULT;
679 : }
680 :
681 0 : for_each_vma(vmi, vma) {
682 : vm_flags_t newflags;
683 :
684 0 : newflags = vma->vm_flags & ~VM_LOCKED_MASK;
685 0 : newflags |= to_add;
686 :
687 : /* Ignore errors */
688 0 : mlock_fixup(&vmi, vma, &prev, vma->vm_start, vma->vm_end,
689 : newflags);
690 0 : cond_resched();
691 : }
692 : out:
693 0 : return 0;
694 : }
695 :
696 0 : SYSCALL_DEFINE1(mlockall, int, flags)
697 : {
698 : unsigned long lock_limit;
699 : int ret;
700 :
701 0 : if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)) ||
702 : flags == MCL_ONFAULT)
703 : return -EINVAL;
704 :
705 0 : if (!can_do_mlock())
706 : return -EPERM;
707 :
708 0 : lock_limit = rlimit(RLIMIT_MEMLOCK);
709 0 : lock_limit >>= PAGE_SHIFT;
710 :
711 0 : if (mmap_write_lock_killable(current->mm))
712 : return -EINTR;
713 :
714 0 : ret = -ENOMEM;
715 0 : if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
716 0 : capable(CAP_IPC_LOCK))
717 0 : ret = apply_mlockall_flags(flags);
718 0 : mmap_write_unlock(current->mm);
719 0 : if (!ret && (flags & MCL_CURRENT))
720 0 : mm_populate(0, TASK_SIZE);
721 :
722 0 : return ret;
723 : }
724 :
725 0 : SYSCALL_DEFINE0(munlockall)
726 : {
727 : int ret;
728 :
729 0 : if (mmap_write_lock_killable(current->mm))
730 : return -EINTR;
731 0 : ret = apply_mlockall_flags(0);
732 0 : mmap_write_unlock(current->mm);
733 0 : return ret;
734 : }
735 :
736 : /*
737 : * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
738 : * shm segments) get accounted against the user_struct instead.
739 : */
740 : static DEFINE_SPINLOCK(shmlock_user_lock);
741 :
742 0 : int user_shm_lock(size_t size, struct ucounts *ucounts)
743 : {
744 : unsigned long lock_limit, locked;
745 : long memlock;
746 0 : int allowed = 0;
747 :
748 0 : locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
749 0 : lock_limit = rlimit(RLIMIT_MEMLOCK);
750 0 : if (lock_limit != RLIM_INFINITY)
751 0 : lock_limit >>= PAGE_SHIFT;
752 0 : spin_lock(&shmlock_user_lock);
753 0 : memlock = inc_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
754 :
755 0 : if ((memlock == LONG_MAX || memlock > lock_limit) && !capable(CAP_IPC_LOCK)) {
756 0 : dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
757 0 : goto out;
758 : }
759 0 : if (!get_ucounts(ucounts)) {
760 0 : dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
761 0 : allowed = 0;
762 0 : goto out;
763 : }
764 : allowed = 1;
765 : out:
766 0 : spin_unlock(&shmlock_user_lock);
767 0 : return allowed;
768 : }
769 :
770 0 : void user_shm_unlock(size_t size, struct ucounts *ucounts)
771 : {
772 0 : spin_lock(&shmlock_user_lock);
773 0 : dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, (size + PAGE_SIZE - 1) >> PAGE_SHIFT);
774 0 : spin_unlock(&shmlock_user_lock);
775 0 : put_ucounts(ucounts);
776 0 : }
|
