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_pvec {
32 : local_lock_t lock;
33 : struct pagevec vec;
34 : };
35 :
36 : static DEFINE_PER_CPU(struct mlock_pvec, mlock_pvec) = {
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 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 :
61 0 : static 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 0 : if (!TestClearPageLRU(page))
65 : return lruvec;
66 :
67 0 : lruvec = folio_lruvec_relock_irq(page_folio(page), lruvec);
68 :
69 0 : 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 0 : if (PageUnevictable(page)) {
77 0 : del_page_from_lru_list(page, lruvec);
78 0 : ClearPageUnevictable(page);
79 0 : add_page_to_lru_list(page, lruvec);
80 0 : __count_vm_events(UNEVICTABLE_PGRESCUED,
81 0 : thp_nr_pages(page));
82 : }
83 : goto out;
84 : }
85 :
86 0 : if (PageUnevictable(page)) {
87 0 : if (PageMlocked(page))
88 0 : page->mlock_count++;
89 : goto out;
90 : }
91 :
92 0 : del_page_from_lru_list(page, lruvec);
93 0 : ClearPageActive(page);
94 0 : SetPageUnevictable(page);
95 0 : page->mlock_count = !!PageMlocked(page);
96 0 : add_page_to_lru_list(page, lruvec);
97 0 : __count_vm_events(UNEVICTABLE_PGCULLED, thp_nr_pages(page));
98 : out:
99 0 : SetPageLRU(page);
100 0 : return lruvec;
101 : }
102 :
103 0 : static struct lruvec *__mlock_new_page(struct page *page, struct lruvec *lruvec)
104 : {
105 : VM_BUG_ON_PAGE(PageLRU(page), page);
106 :
107 0 : lruvec = folio_lruvec_relock_irq(page_folio(page), lruvec);
108 :
109 : /* As above, this is a little surprising, but possible */
110 0 : if (unlikely(page_evictable(page)))
111 : goto out;
112 :
113 0 : SetPageUnevictable(page);
114 0 : page->mlock_count = !!PageMlocked(page);
115 0 : __count_vm_events(UNEVICTABLE_PGCULLED, thp_nr_pages(page));
116 : out:
117 0 : add_page_to_lru_list(page, lruvec);
118 0 : SetPageLRU(page);
119 0 : return lruvec;
120 : }
121 :
122 0 : static struct lruvec *__munlock_page(struct page *page, struct lruvec *lruvec)
123 : {
124 0 : int nr_pages = thp_nr_pages(page);
125 0 : bool isolated = false;
126 :
127 0 : if (!TestClearPageLRU(page))
128 : goto munlock;
129 :
130 0 : isolated = true;
131 0 : lruvec = folio_lruvec_relock_irq(page_folio(page), lruvec);
132 :
133 0 : if (PageUnevictable(page)) {
134 : /* Then mlock_count is maintained, but might undercount */
135 0 : if (page->mlock_count)
136 0 : page->mlock_count--;
137 0 : if (page->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 (TestClearPageMlocked(page)) {
144 0 : __mod_zone_page_state(page_zone(page), NR_MLOCK, -nr_pages);
145 0 : if (isolated || !PageUnevictable(page))
146 0 : __count_vm_events(UNEVICTABLE_PGMUNLOCKED, nr_pages);
147 : else
148 0 : __count_vm_events(UNEVICTABLE_PGSTRANDED, nr_pages);
149 : }
150 :
151 : /* page_evictable() has to be checked *after* clearing Mlocked */
152 0 : if (isolated && PageUnevictable(page) && page_evictable(page)) {
153 0 : del_page_from_lru_list(page, lruvec);
154 0 : ClearPageUnevictable(page);
155 0 : add_page_to_lru_list(page, lruvec);
156 0 : __count_vm_events(UNEVICTABLE_PGRESCUED, nr_pages);
157 : }
158 : out:
159 0 : if (isolated)
160 : SetPageLRU(page);
161 0 : 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
169 : static inline struct page *mlock_lru(struct page *page)
170 : {
171 0 : return (struct page *)((unsigned long)page + LRU_PAGE);
172 : }
173 :
174 : static inline struct page *mlock_new(struct page *page)
175 : {
176 0 : 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 : */
186 0 : static void mlock_pagevec(struct pagevec *pvec)
187 : {
188 0 : struct lruvec *lruvec = NULL;
189 : unsigned long mlock;
190 : struct page *page;
191 : int i;
192 :
193 0 : for (i = 0; i < pagevec_count(pvec); i++) {
194 0 : page = pvec->pages[i];
195 0 : mlock = (unsigned long)page & (LRU_PAGE | NEW_PAGE);
196 0 : page = (struct page *)((unsigned long)page - mlock);
197 0 : pvec->pages[i] = page;
198 :
199 0 : if (mlock & LRU_PAGE)
200 0 : lruvec = __mlock_page(page, lruvec);
201 0 : else if (mlock & NEW_PAGE)
202 0 : lruvec = __mlock_new_page(page, lruvec);
203 : else
204 0 : lruvec = __munlock_page(page, lruvec);
205 : }
206 :
207 0 : if (lruvec)
208 0 : unlock_page_lruvec_irq(lruvec);
209 0 : release_pages(pvec->pages, pvec->nr);
210 0 : pagevec_reinit(pvec);
211 0 : }
212 :
213 0 : void mlock_page_drain_local(void)
214 : {
215 : struct pagevec *pvec;
216 :
217 0 : local_lock(&mlock_pvec.lock);
218 0 : pvec = this_cpu_ptr(&mlock_pvec.vec);
219 0 : if (pagevec_count(pvec))
220 0 : mlock_pagevec(pvec);
221 0 : local_unlock(&mlock_pvec.lock);
222 0 : }
223 :
224 0 : void mlock_page_drain_remote(int cpu)
225 : {
226 : struct pagevec *pvec;
227 :
228 0 : WARN_ON_ONCE(cpu_online(cpu));
229 0 : pvec = &per_cpu(mlock_pvec.vec, cpu);
230 0 : if (pagevec_count(pvec))
231 0 : mlock_pagevec(pvec);
232 0 : }
233 :
234 0 : bool need_mlock_page_drain(int cpu)
235 : {
236 0 : 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 : */
243 0 : void mlock_folio(struct folio *folio)
244 : {
245 : struct pagevec *pvec;
246 :
247 0 : local_lock(&mlock_pvec.lock);
248 0 : pvec = this_cpu_ptr(&mlock_pvec.vec);
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 (!pagevec_add(pvec, mlock_lru(&folio->page)) ||
259 0 : folio_test_large(folio) || lru_cache_disabled())
260 0 : mlock_pagevec(pvec);
261 0 : local_unlock(&mlock_pvec.lock);
262 0 : }
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 : */
268 0 : void mlock_new_page(struct page *page)
269 : {
270 : struct pagevec *pvec;
271 0 : int nr_pages = thp_nr_pages(page);
272 :
273 0 : local_lock(&mlock_pvec.lock);
274 0 : pvec = this_cpu_ptr(&mlock_pvec.vec);
275 0 : SetPageMlocked(page);
276 0 : mod_zone_page_state(page_zone(page), NR_MLOCK, nr_pages);
277 0 : __count_vm_events(UNEVICTABLE_PGMLOCKED, nr_pages);
278 :
279 0 : get_page(page);
280 0 : if (!pagevec_add(pvec, mlock_new(page)) ||
281 0 : PageHead(page) || lru_cache_disabled())
282 0 : mlock_pagevec(pvec);
283 0 : local_unlock(&mlock_pvec.lock);
284 0 : }
285 :
286 : /**
287 : * munlock_page - munlock a page
288 : * @page: page to be munlocked, either a normal page or a THP head.
289 : */
290 0 : void munlock_page(struct page *page)
291 : {
292 : struct pagevec *pvec;
293 :
294 0 : local_lock(&mlock_pvec.lock);
295 0 : 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 0 : get_page(page);
302 0 : if (!pagevec_add(pvec, page) ||
303 0 : PageHead(page) || lru_cache_disabled())
304 0 : mlock_pagevec(pvec);
305 0 : local_unlock(&mlock_pvec.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 page *page;
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 : 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 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 : page = vm_normal_page(vma, addr, *pte);
336 0 : if (!page)
337 0 : continue;
338 0 : if (PageTransCompound(page))
339 : continue;
340 0 : if (vma->vm_flags & VM_LOCKED)
341 0 : mlock_folio(page_folio(page));
342 : else
343 0 : munlock_page(page);
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_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 0 : if (newflags & VM_LOCKED)
382 0 : newflags |= VM_IO;
383 0 : WRITE_ONCE(vma->vm_flags, 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 0 : WRITE_ONCE(vma->vm_flags, 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 vm_area_struct *vma, struct vm_area_struct **prev,
405 : unsigned long start, unsigned long end, vm_flags_t newflags)
406 : {
407 0 : struct mm_struct *mm = vma->vm_mm;
408 : pgoff_t pgoff;
409 : int nr_pages;
410 0 : int ret = 0;
411 0 : vm_flags_t oldflags = vma->vm_flags;
412 :
413 0 : if (newflags == oldflags || (oldflags & VM_SPECIAL) ||
414 0 : is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm) ||
415 0 : 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 0 : pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
420 0 : *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 0 : if (*prev) {
424 : vma = *prev;
425 : goto success;
426 : }
427 :
428 0 : if (start != vma->vm_start) {
429 0 : ret = split_vma(mm, vma, start, 1);
430 0 : if (ret)
431 : goto out;
432 : }
433 :
434 0 : if (end != vma->vm_end) {
435 0 : ret = split_vma(mm, vma, end, 0);
436 0 : if (ret)
437 : goto out;
438 : }
439 :
440 : success:
441 : /*
442 : * Keep track of amount of locked VM.
443 : */
444 0 : nr_pages = (end - start) >> PAGE_SHIFT;
445 0 : if (!(newflags & VM_LOCKED))
446 0 : nr_pages = -nr_pages;
447 0 : else if (oldflags & VM_LOCKED)
448 0 : nr_pages = 0;
449 0 : 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 0 : if ((newflags & VM_LOCKED) && (oldflags & VM_LOCKED)) {
458 : /* No work to do, and mlocking twice would be wrong */
459 0 : vma->vm_flags = newflags;
460 : } else {
461 0 : mlock_vma_pages_range(vma, start, end, newflags);
462 : }
463 : out:
464 0 : *prev = vma;
465 0 : return ret;
466 : }
467 :
468 0 : static 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 :
475 : VM_BUG_ON(offset_in_page(start));
476 : VM_BUG_ON(len != PAGE_ALIGN(len));
477 0 : end = start + len;
478 0 : if (end < start)
479 : return -EINVAL;
480 0 : if (end == start)
481 : return 0;
482 0 : vma = find_vma(current->mm, start);
483 0 : if (!vma || vma->vm_start > start)
484 : return -ENOMEM;
485 :
486 0 : prev = vma->vm_prev;
487 0 : if (start > vma->vm_start)
488 0 : prev = vma;
489 :
490 : for (nstart = start ; ; ) {
491 0 : vm_flags_t newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
492 :
493 0 : newflags |= flags;
494 :
495 : /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
496 0 : tmp = vma->vm_end;
497 0 : if (tmp > end)
498 0 : tmp = end;
499 0 : error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
500 0 : if (error)
501 : break;
502 0 : nstart = tmp;
503 0 : if (nstart < prev->vm_end)
504 0 : nstart = prev->vm_end;
505 0 : if (nstart >= end)
506 : break;
507 :
508 0 : vma = prev->vm_next;
509 0 : if (!vma || vma->vm_start != nstart) {
510 : error = -ENOMEM;
511 : break;
512 : }
513 : }
514 : return error;
515 : }
516 :
517 : /*
518 : * Go through vma areas and sum size of mlocked
519 : * vma pages, as return value.
520 : * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT)
521 : * is also counted.
522 : * Return value: previously mlocked page counts
523 : */
524 0 : static unsigned long count_mm_mlocked_page_nr(struct mm_struct *mm,
525 : unsigned long start, size_t len)
526 : {
527 : struct vm_area_struct *vma;
528 0 : unsigned long count = 0;
529 :
530 0 : if (mm == NULL)
531 0 : mm = current->mm;
532 :
533 0 : vma = find_vma(mm, start);
534 0 : if (vma == NULL)
535 : return 0;
536 :
537 0 : for (; vma ; vma = vma->vm_next) {
538 0 : if (start >= vma->vm_end)
539 0 : continue;
540 0 : if (start + len <= vma->vm_start)
541 : break;
542 0 : if (vma->vm_flags & VM_LOCKED) {
543 0 : if (start > vma->vm_start)
544 0 : count -= (start - vma->vm_start);
545 0 : if (start + len < vma->vm_end) {
546 0 : count += start + len - vma->vm_start;
547 0 : break;
548 : }
549 0 : count += vma->vm_end - vma->vm_start;
550 : }
551 : }
552 :
553 0 : return count >> PAGE_SHIFT;
554 : }
555 :
556 : /*
557 : * convert get_user_pages() return value to posix mlock() error
558 : */
559 : static int __mlock_posix_error_return(long retval)
560 : {
561 0 : if (retval == -EFAULT)
562 : retval = -ENOMEM;
563 0 : else if (retval == -ENOMEM)
564 0 : retval = -EAGAIN;
565 0 : return retval;
566 : }
567 :
568 0 : static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
569 : {
570 : unsigned long locked;
571 : unsigned long lock_limit;
572 0 : int error = -ENOMEM;
573 :
574 0 : start = untagged_addr(start);
575 :
576 0 : if (!can_do_mlock())
577 : return -EPERM;
578 :
579 0 : len = PAGE_ALIGN(len + (offset_in_page(start)));
580 0 : start &= PAGE_MASK;
581 :
582 0 : lock_limit = rlimit(RLIMIT_MEMLOCK);
583 0 : lock_limit >>= PAGE_SHIFT;
584 0 : locked = len >> PAGE_SHIFT;
585 :
586 0 : if (mmap_write_lock_killable(current->mm))
587 : return -EINTR;
588 :
589 0 : locked += current->mm->locked_vm;
590 0 : if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) {
591 : /*
592 : * It is possible that the regions requested intersect with
593 : * previously mlocked areas, that part area in "mm->locked_vm"
594 : * should not be counted to new mlock increment count. So check
595 : * and adjust locked count if necessary.
596 : */
597 0 : locked -= count_mm_mlocked_page_nr(current->mm,
598 : start, len);
599 : }
600 :
601 : /* check against resource limits */
602 0 : if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
603 0 : error = apply_vma_lock_flags(start, len, flags);
604 :
605 0 : mmap_write_unlock(current->mm);
606 0 : if (error)
607 : return error;
608 :
609 0 : error = __mm_populate(start, len, 0);
610 0 : if (error)
611 0 : return __mlock_posix_error_return(error);
612 : return 0;
613 : }
614 :
615 0 : SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
616 : {
617 0 : return do_mlock(start, len, VM_LOCKED);
618 : }
619 :
620 0 : SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
621 : {
622 0 : vm_flags_t vm_flags = VM_LOCKED;
623 :
624 0 : if (flags & ~MLOCK_ONFAULT)
625 : return -EINVAL;
626 :
627 0 : if (flags & MLOCK_ONFAULT)
628 0 : vm_flags |= VM_LOCKONFAULT;
629 :
630 0 : return do_mlock(start, len, vm_flags);
631 : }
632 :
633 0 : SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
634 : {
635 : int ret;
636 :
637 0 : start = untagged_addr(start);
638 :
639 0 : len = PAGE_ALIGN(len + (offset_in_page(start)));
640 0 : start &= PAGE_MASK;
641 :
642 0 : if (mmap_write_lock_killable(current->mm))
643 : return -EINTR;
644 0 : ret = apply_vma_lock_flags(start, len, 0);
645 0 : mmap_write_unlock(current->mm);
646 :
647 0 : return ret;
648 : }
649 :
650 : /*
651 : * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
652 : * and translate into the appropriate modifications to mm->def_flags and/or the
653 : * flags for all current VMAs.
654 : *
655 : * There are a couple of subtleties with this. If mlockall() is called multiple
656 : * times with different flags, the values do not necessarily stack. If mlockall
657 : * is called once including the MCL_FUTURE flag and then a second time without
658 : * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
659 : */
660 0 : static int apply_mlockall_flags(int flags)
661 : {
662 0 : struct vm_area_struct *vma, *prev = NULL;
663 0 : vm_flags_t to_add = 0;
664 :
665 0 : current->mm->def_flags &= VM_LOCKED_CLEAR_MASK;
666 0 : if (flags & MCL_FUTURE) {
667 0 : current->mm->def_flags |= VM_LOCKED;
668 :
669 0 : if (flags & MCL_ONFAULT)
670 0 : current->mm->def_flags |= VM_LOCKONFAULT;
671 :
672 0 : if (!(flags & MCL_CURRENT))
673 : goto out;
674 : }
675 :
676 0 : if (flags & MCL_CURRENT) {
677 0 : to_add |= VM_LOCKED;
678 0 : if (flags & MCL_ONFAULT)
679 0 : to_add |= VM_LOCKONFAULT;
680 : }
681 :
682 0 : for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
683 : vm_flags_t newflags;
684 :
685 0 : newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
686 0 : newflags |= to_add;
687 :
688 : /* Ignore errors */
689 0 : mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, 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 : }
|
