]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - mm/mlock.c
Merge tag 'for-4.15-rc7-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave...
[mirror_ubuntu-bionic-kernel.git] / mm / mlock.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
1da177e4
LT
2/*
3 * linux/mm/mlock.c
4 *
5 * (C) Copyright 1995 Linus Torvalds
6 * (C) Copyright 2002 Christoph Hellwig
7 */
8
c59ede7b 9#include <linux/capability.h>
1da177e4
LT
10#include <linux/mman.h>
11#include <linux/mm.h>
8703e8a4 12#include <linux/sched/user.h>
b291f000
NP
13#include <linux/swap.h>
14#include <linux/swapops.h>
15#include <linux/pagemap.h>
7225522b 16#include <linux/pagevec.h>
1da177e4
LT
17#include <linux/mempolicy.h>
18#include <linux/syscalls.h>
e8edc6e0 19#include <linux/sched.h>
b95f1b31 20#include <linux/export.h>
b291f000
NP
21#include <linux/rmap.h>
22#include <linux/mmzone.h>
23#include <linux/hugetlb.h>
7225522b
VB
24#include <linux/memcontrol.h>
25#include <linux/mm_inline.h>
b291f000
NP
26
27#include "internal.h"
1da177e4 28
7f43add4 29bool can_do_mlock(void)
e8edc6e0 30{
59e99e5b 31 if (rlimit(RLIMIT_MEMLOCK) != 0)
7f43add4 32 return true;
a5a6579d 33 if (capable(CAP_IPC_LOCK))
7f43add4
WX
34 return true;
35 return false;
e8edc6e0
AD
36}
37EXPORT_SYMBOL(can_do_mlock);
1da177e4 38
b291f000
NP
39/*
40 * Mlocked pages are marked with PageMlocked() flag for efficient testing
41 * in vmscan and, possibly, the fault path; and to support semi-accurate
42 * statistics.
43 *
44 * An mlocked page [PageMlocked(page)] is unevictable. As such, it will
45 * be placed on the LRU "unevictable" list, rather than the [in]active lists.
46 * The unevictable list is an LRU sibling list to the [in]active lists.
47 * PageUnevictable is set to indicate the unevictable state.
48 *
49 * When lazy mlocking via vmscan, it is important to ensure that the
50 * vma's VM_LOCKED status is not concurrently being modified, otherwise we
51 * may have mlocked a page that is being munlocked. So lazy mlock must take
52 * the mmap_sem for read, and verify that the vma really is locked
53 * (see mm/rmap.c).
54 */
55
56/*
57 * LRU accounting for clear_page_mlock()
58 */
e6c509f8 59void clear_page_mlock(struct page *page)
b291f000 60{
e6c509f8 61 if (!TestClearPageMlocked(page))
b291f000 62 return;
b291f000 63
8449d21f
DR
64 mod_zone_page_state(page_zone(page), NR_MLOCK,
65 -hpage_nr_pages(page));
5344b7e6 66 count_vm_event(UNEVICTABLE_PGCLEARED);
b291f000
NP
67 if (!isolate_lru_page(page)) {
68 putback_lru_page(page);
69 } else {
70 /*
8891d6da 71 * We lost the race. the page already moved to evictable list.
b291f000 72 */
8891d6da 73 if (PageUnevictable(page))
5344b7e6 74 count_vm_event(UNEVICTABLE_PGSTRANDED);
b291f000
NP
75 }
76}
77
78/*
79 * Mark page as mlocked if not already.
80 * If page on LRU, isolate and putback to move to unevictable list.
81 */
82void mlock_vma_page(struct page *page)
83{
57e68e9c 84 /* Serialize with page migration */
b291f000
NP
85 BUG_ON(!PageLocked(page));
86
e90309c9
KS
87 VM_BUG_ON_PAGE(PageTail(page), page);
88 VM_BUG_ON_PAGE(PageCompound(page) && PageDoubleMap(page), page);
89
5344b7e6 90 if (!TestSetPageMlocked(page)) {
8449d21f
DR
91 mod_zone_page_state(page_zone(page), NR_MLOCK,
92 hpage_nr_pages(page));
5344b7e6
NP
93 count_vm_event(UNEVICTABLE_PGMLOCKED);
94 if (!isolate_lru_page(page))
95 putback_lru_page(page);
96 }
b291f000
NP
97}
98
01cc2e58
VB
99/*
100 * Isolate a page from LRU with optional get_page() pin.
101 * Assumes lru_lock already held and page already pinned.
102 */
103static bool __munlock_isolate_lru_page(struct page *page, bool getpage)
104{
105 if (PageLRU(page)) {
106 struct lruvec *lruvec;
107
599d0c95 108 lruvec = mem_cgroup_page_lruvec(page, page_pgdat(page));
01cc2e58
VB
109 if (getpage)
110 get_page(page);
111 ClearPageLRU(page);
112 del_page_from_lru_list(page, lruvec, page_lru(page));
113 return true;
114 }
115
116 return false;
117}
118
7225522b
VB
119/*
120 * Finish munlock after successful page isolation
121 *
122 * Page must be locked. This is a wrapper for try_to_munlock()
123 * and putback_lru_page() with munlock accounting.
124 */
125static void __munlock_isolated_page(struct page *page)
126{
7225522b
VB
127 /*
128 * Optimization: if the page was mapped just once, that's our mapping
129 * and we don't need to check all the other vmas.
130 */
131 if (page_mapcount(page) > 1)
192d7232 132 try_to_munlock(page);
7225522b
VB
133
134 /* Did try_to_unlock() succeed or punt? */
192d7232 135 if (!PageMlocked(page))
7225522b
VB
136 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
137
138 putback_lru_page(page);
139}
140
141/*
142 * Accounting for page isolation fail during munlock
143 *
144 * Performs accounting when page isolation fails in munlock. There is nothing
145 * else to do because it means some other task has already removed the page
146 * from the LRU. putback_lru_page() will take care of removing the page from
147 * the unevictable list, if necessary. vmscan [page_referenced()] will move
148 * the page back to the unevictable list if some other vma has it mlocked.
149 */
150static void __munlock_isolation_failed(struct page *page)
151{
152 if (PageUnevictable(page))
01cc2e58 153 __count_vm_event(UNEVICTABLE_PGSTRANDED);
7225522b 154 else
01cc2e58 155 __count_vm_event(UNEVICTABLE_PGMUNLOCKED);
7225522b
VB
156}
157
6927c1dd
LS
158/**
159 * munlock_vma_page - munlock a vma page
c424be1c
VB
160 * @page - page to be unlocked, either a normal page or THP page head
161 *
162 * returns the size of the page as a page mask (0 for normal page,
163 * HPAGE_PMD_NR - 1 for THP head page)
b291f000 164 *
6927c1dd
LS
165 * called from munlock()/munmap() path with page supposedly on the LRU.
166 * When we munlock a page, because the vma where we found the page is being
167 * munlock()ed or munmap()ed, we want to check whether other vmas hold the
168 * page locked so that we can leave it on the unevictable lru list and not
169 * bother vmscan with it. However, to walk the page's rmap list in
170 * try_to_munlock() we must isolate the page from the LRU. If some other
171 * task has removed the page from the LRU, we won't be able to do that.
172 * So we clear the PageMlocked as we might not get another chance. If we
173 * can't isolate the page, we leave it for putback_lru_page() and vmscan
174 * [page_referenced()/try_to_unmap()] to deal with.
b291f000 175 */
ff6a6da6 176unsigned int munlock_vma_page(struct page *page)
b291f000 177{
7162a1e8 178 int nr_pages;
01cc2e58 179 struct zone *zone = page_zone(page);
ff6a6da6 180
57e68e9c 181 /* For try_to_munlock() and to serialize with page migration */
b291f000
NP
182 BUG_ON(!PageLocked(page));
183
e90309c9
KS
184 VM_BUG_ON_PAGE(PageTail(page), page);
185
c424be1c 186 /*
01cc2e58
VB
187 * Serialize with any parallel __split_huge_page_refcount() which
188 * might otherwise copy PageMlocked to part of the tail pages before
189 * we clear it in the head page. It also stabilizes hpage_nr_pages().
c424be1c 190 */
a52633d8 191 spin_lock_irq(zone_lru_lock(zone));
01cc2e58 192
655548bf
KS
193 if (!TestClearPageMlocked(page)) {
194 /* Potentially, PTE-mapped THP: do not skip the rest PTEs */
195 nr_pages = 1;
01cc2e58 196 goto unlock_out;
655548bf 197 }
01cc2e58 198
655548bf 199 nr_pages = hpage_nr_pages(page);
01cc2e58
VB
200 __mod_zone_page_state(zone, NR_MLOCK, -nr_pages);
201
202 if (__munlock_isolate_lru_page(page, true)) {
a52633d8 203 spin_unlock_irq(zone_lru_lock(zone));
01cc2e58
VB
204 __munlock_isolated_page(page);
205 goto out;
206 }
207 __munlock_isolation_failed(page);
208
209unlock_out:
a52633d8 210 spin_unlock_irq(zone_lru_lock(zone));
01cc2e58
VB
211
212out:
c424be1c 213 return nr_pages - 1;
b291f000
NP
214}
215
9978ad58
LS
216/*
217 * convert get_user_pages() return value to posix mlock() error
218 */
219static int __mlock_posix_error_return(long retval)
220{
221 if (retval == -EFAULT)
222 retval = -ENOMEM;
223 else if (retval == -ENOMEM)
224 retval = -EAGAIN;
225 return retval;
b291f000
NP
226}
227
56afe477
VB
228/*
229 * Prepare page for fast batched LRU putback via putback_lru_evictable_pagevec()
230 *
231 * The fast path is available only for evictable pages with single mapping.
232 * Then we can bypass the per-cpu pvec and get better performance.
233 * when mapcount > 1 we need try_to_munlock() which can fail.
234 * when !page_evictable(), we need the full redo logic of putback_lru_page to
235 * avoid leaving evictable page in unevictable list.
236 *
237 * In case of success, @page is added to @pvec and @pgrescued is incremented
238 * in case that the page was previously unevictable. @page is also unlocked.
239 */
240static bool __putback_lru_fast_prepare(struct page *page, struct pagevec *pvec,
241 int *pgrescued)
242{
309381fe
SL
243 VM_BUG_ON_PAGE(PageLRU(page), page);
244 VM_BUG_ON_PAGE(!PageLocked(page), page);
56afe477
VB
245
246 if (page_mapcount(page) <= 1 && page_evictable(page)) {
247 pagevec_add(pvec, page);
248 if (TestClearPageUnevictable(page))
249 (*pgrescued)++;
250 unlock_page(page);
251 return true;
252 }
253
254 return false;
255}
256
257/*
258 * Putback multiple evictable pages to the LRU
259 *
260 * Batched putback of evictable pages that bypasses the per-cpu pvec. Some of
261 * the pages might have meanwhile become unevictable but that is OK.
262 */
263static void __putback_lru_fast(struct pagevec *pvec, int pgrescued)
264{
265 count_vm_events(UNEVICTABLE_PGMUNLOCKED, pagevec_count(pvec));
266 /*
267 *__pagevec_lru_add() calls release_pages() so we don't call
268 * put_page() explicitly
269 */
270 __pagevec_lru_add(pvec);
271 count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued);
272}
273
7225522b
VB
274/*
275 * Munlock a batch of pages from the same zone
276 *
277 * The work is split to two main phases. First phase clears the Mlocked flag
278 * and attempts to isolate the pages, all under a single zone lru lock.
279 * The second phase finishes the munlock only for pages where isolation
280 * succeeded.
281 *
7a8010cd 282 * Note that the pagevec may be modified during the process.
7225522b
VB
283 */
284static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone)
285{
286 int i;
287 int nr = pagevec_count(pvec);
70feee0e 288 int delta_munlocked = -nr;
56afe477
VB
289 struct pagevec pvec_putback;
290 int pgrescued = 0;
7225522b 291
86679820 292 pagevec_init(&pvec_putback);
3b25df93 293
7225522b 294 /* Phase 1: page isolation */
a52633d8 295 spin_lock_irq(zone_lru_lock(zone));
7225522b
VB
296 for (i = 0; i < nr; i++) {
297 struct page *page = pvec->pages[i];
298
299 if (TestClearPageMlocked(page)) {
7225522b 300 /*
01cc2e58
VB
301 * We already have pin from follow_page_mask()
302 * so we can spare the get_page() here.
7225522b 303 */
01cc2e58
VB
304 if (__munlock_isolate_lru_page(page, false))
305 continue;
306 else
307 __munlock_isolation_failed(page);
70feee0e
YX
308 } else {
309 delta_munlocked++;
7225522b 310 }
01cc2e58
VB
311
312 /*
313 * We won't be munlocking this page in the next phase
314 * but we still need to release the follow_page_mask()
315 * pin. We cannot do it under lru_lock however. If it's
316 * the last pin, __page_cache_release() would deadlock.
317 */
318 pagevec_add(&pvec_putback, pvec->pages[i]);
319 pvec->pages[i] = NULL;
7225522b 320 }
1ebb7cc6 321 __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked);
a52633d8 322 spin_unlock_irq(zone_lru_lock(zone));
7225522b 323
3b25df93
VB
324 /* Now we can release pins of pages that we are not munlocking */
325 pagevec_release(&pvec_putback);
326
56afe477 327 /* Phase 2: page munlock */
7225522b
VB
328 for (i = 0; i < nr; i++) {
329 struct page *page = pvec->pages[i];
330
331 if (page) {
332 lock_page(page);
56afe477
VB
333 if (!__putback_lru_fast_prepare(page, &pvec_putback,
334 &pgrescued)) {
5b40998a
VB
335 /*
336 * Slow path. We don't want to lose the last
337 * pin before unlock_page()
338 */
339 get_page(page); /* for putback_lru_page() */
56afe477
VB
340 __munlock_isolated_page(page);
341 unlock_page(page);
5b40998a 342 put_page(page); /* from follow_page_mask() */
56afe477 343 }
7225522b
VB
344 }
345 }
56afe477 346
5b40998a
VB
347 /*
348 * Phase 3: page putback for pages that qualified for the fast path
349 * This will also call put_page() to return pin from follow_page_mask()
350 */
56afe477
VB
351 if (pagevec_count(&pvec_putback))
352 __putback_lru_fast(&pvec_putback, pgrescued);
7a8010cd
VB
353}
354
355/*
356 * Fill up pagevec for __munlock_pagevec using pte walk
357 *
358 * The function expects that the struct page corresponding to @start address is
359 * a non-TPH page already pinned and in the @pvec, and that it belongs to @zone.
360 *
361 * The rest of @pvec is filled by subsequent pages within the same pmd and same
362 * zone, as long as the pte's are present and vm_normal_page() succeeds. These
363 * pages also get pinned.
364 *
365 * Returns the address of the next page that should be scanned. This equals
366 * @start + PAGE_SIZE when no page could be added by the pte walk.
367 */
368static unsigned long __munlock_pagevec_fill(struct pagevec *pvec,
9472f23c
JK
369 struct vm_area_struct *vma, struct zone *zone,
370 unsigned long start, unsigned long end)
7a8010cd
VB
371{
372 pte_t *pte;
373 spinlock_t *ptl;
374
375 /*
376 * Initialize pte walk starting at the already pinned page where we
eadb41ae
VB
377 * are sure that there is a pte, as it was pinned under the same
378 * mmap_sem write op.
7a8010cd
VB
379 */
380 pte = get_locked_pte(vma->vm_mm, start, &ptl);
eadb41ae
VB
381 /* Make sure we do not cross the page table boundary */
382 end = pgd_addr_end(start, end);
c2febafc 383 end = p4d_addr_end(start, end);
eadb41ae
VB
384 end = pud_addr_end(start, end);
385 end = pmd_addr_end(start, end);
7a8010cd
VB
386
387 /* The page next to the pinned page is the first we will try to get */
388 start += PAGE_SIZE;
389 while (start < end) {
390 struct page *page = NULL;
391 pte++;
392 if (pte_present(*pte))
393 page = vm_normal_page(vma, start, *pte);
394 /*
395 * Break if page could not be obtained or the page's node+zone does not
396 * match
397 */
9472f23c 398 if (!page || page_zone(page) != zone)
7a8010cd 399 break;
56afe477 400
e90309c9
KS
401 /*
402 * Do not use pagevec for PTE-mapped THP,
403 * munlock_vma_pages_range() will handle them.
404 */
405 if (PageTransCompound(page))
406 break;
407
7a8010cd
VB
408 get_page(page);
409 /*
410 * Increase the address that will be returned *before* the
411 * eventual break due to pvec becoming full by adding the page
412 */
413 start += PAGE_SIZE;
414 if (pagevec_add(pvec, page) == 0)
415 break;
416 }
417 pte_unmap_unlock(pte, ptl);
418 return start;
7225522b
VB
419}
420
b291f000 421/*
ba470de4
RR
422 * munlock_vma_pages_range() - munlock all pages in the vma range.'
423 * @vma - vma containing range to be munlock()ed.
424 * @start - start address in @vma of the range
425 * @end - end of range in @vma.
426 *
427 * For mremap(), munmap() and exit().
428 *
429 * Called with @vma VM_LOCKED.
430 *
431 * Returns with VM_LOCKED cleared. Callers must be prepared to
432 * deal with this.
433 *
434 * We don't save and restore VM_LOCKED here because pages are
435 * still on lru. In unmap path, pages might be scanned by reclaim
436 * and re-mlocked by try_to_{munlock|unmap} before we unmap and
437 * free them. This will result in freeing mlocked pages.
b291f000 438 */
ba470de4 439void munlock_vma_pages_range(struct vm_area_struct *vma,
408e82b7 440 unsigned long start, unsigned long end)
b291f000 441{
de60f5f1 442 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
408e82b7 443
ff6a6da6 444 while (start < end) {
ab7a5af7 445 struct page *page;
6ebb4a1b 446 unsigned int page_mask = 0;
c424be1c 447 unsigned long page_increm;
7a8010cd
VB
448 struct pagevec pvec;
449 struct zone *zone;
ff6a6da6 450
86679820 451 pagevec_init(&pvec);
6e919717
HD
452 /*
453 * Although FOLL_DUMP is intended for get_dump_page(),
454 * it just so happens that its special treatment of the
455 * ZERO_PAGE (returning an error instead of doing get_page)
456 * suits munlock very well (and if somehow an abnormal page
457 * has sneaked into the range, we won't oops here: great).
458 */
6ebb4a1b 459 page = follow_page(vma, start, FOLL_GET | FOLL_DUMP);
7a8010cd 460
e90309c9
KS
461 if (page && !IS_ERR(page)) {
462 if (PageTransTail(page)) {
463 VM_BUG_ON_PAGE(PageMlocked(page), page);
464 put_page(page); /* follow_page_mask() */
465 } else if (PageTransHuge(page)) {
466 lock_page(page);
467 /*
468 * Any THP page found by follow_page_mask() may
469 * have gotten split before reaching
6ebb4a1b
KS
470 * munlock_vma_page(), so we need to compute
471 * the page_mask here instead.
e90309c9
KS
472 */
473 page_mask = munlock_vma_page(page);
474 unlock_page(page);
475 put_page(page); /* follow_page_mask() */
476 } else {
477 /*
478 * Non-huge pages are handled in batches via
479 * pagevec. The pin from follow_page_mask()
480 * prevents them from collapsing by THP.
481 */
482 pagevec_add(&pvec, page);
483 zone = page_zone(page);
7a8010cd 484
e90309c9
KS
485 /*
486 * Try to fill the rest of pagevec using fast
487 * pte walk. This will also update start to
488 * the next page to process. Then munlock the
489 * pagevec.
490 */
491 start = __munlock_pagevec_fill(&pvec, vma,
9472f23c 492 zone, start, end);
e90309c9
KS
493 __munlock_pagevec(&pvec, zone);
494 goto next;
495 }
408e82b7 496 }
c424be1c 497 page_increm = 1 + page_mask;
ff6a6da6 498 start += page_increm * PAGE_SIZE;
7a8010cd 499next:
408e82b7
HD
500 cond_resched();
501 }
b291f000
NP
502}
503
504/*
505 * mlock_fixup - handle mlock[all]/munlock[all] requests.
506 *
507 * Filters out "special" vmas -- VM_LOCKED never gets set for these, and
508 * munlock is a no-op. However, for some special vmas, we go ahead and
cea10a19 509 * populate the ptes.
b291f000
NP
510 *
511 * For vmas that pass the filters, merge/split as appropriate.
512 */
1da177e4 513static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
ca16d140 514 unsigned long start, unsigned long end, vm_flags_t newflags)
1da177e4 515{
b291f000 516 struct mm_struct *mm = vma->vm_mm;
1da177e4 517 pgoff_t pgoff;
b291f000 518 int nr_pages;
1da177e4 519 int ret = 0;
ca16d140 520 int lock = !!(newflags & VM_LOCKED);
b155b4fd 521 vm_flags_t old_flags = vma->vm_flags;
1da177e4 522
fed067da 523 if (newflags == vma->vm_flags || (vma->vm_flags & VM_SPECIAL) ||
31db58b3 524 is_vm_hugetlb_page(vma) || vma == get_gate_vma(current->mm))
b0f205c2
EM
525 /* don't set VM_LOCKED or VM_LOCKONFAULT and don't count */
526 goto out;
b291f000 527
1da177e4
LT
528 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
529 *prev = vma_merge(mm, *prev, start, end, newflags, vma->anon_vma,
19a809af
AA
530 vma->vm_file, pgoff, vma_policy(vma),
531 vma->vm_userfaultfd_ctx);
1da177e4
LT
532 if (*prev) {
533 vma = *prev;
534 goto success;
535 }
536
1da177e4
LT
537 if (start != vma->vm_start) {
538 ret = split_vma(mm, vma, start, 1);
539 if (ret)
540 goto out;
541 }
542
543 if (end != vma->vm_end) {
544 ret = split_vma(mm, vma, end, 0);
545 if (ret)
546 goto out;
547 }
548
549success:
b291f000
NP
550 /*
551 * Keep track of amount of locked VM.
552 */
553 nr_pages = (end - start) >> PAGE_SHIFT;
554 if (!lock)
555 nr_pages = -nr_pages;
b155b4fd
SG
556 else if (old_flags & VM_LOCKED)
557 nr_pages = 0;
b291f000
NP
558 mm->locked_vm += nr_pages;
559
1da177e4
LT
560 /*
561 * vm_flags is protected by the mmap_sem held in write mode.
562 * It's okay if try_to_unmap_one unmaps a page just after we
fc05f566 563 * set VM_LOCKED, populate_vma_page_range will bring it back.
1da177e4 564 */
1da177e4 565
fed067da 566 if (lock)
408e82b7 567 vma->vm_flags = newflags;
fed067da 568 else
408e82b7 569 munlock_vma_pages_range(vma, start, end);
1da177e4 570
1da177e4 571out:
b291f000 572 *prev = vma;
1da177e4
LT
573 return ret;
574}
575
1aab92ec
EM
576static int apply_vma_lock_flags(unsigned long start, size_t len,
577 vm_flags_t flags)
1da177e4
LT
578{
579 unsigned long nstart, end, tmp;
580 struct vm_area_struct * vma, * prev;
581 int error;
582
8fd9e488 583 VM_BUG_ON(offset_in_page(start));
fed067da 584 VM_BUG_ON(len != PAGE_ALIGN(len));
1da177e4
LT
585 end = start + len;
586 if (end < start)
587 return -EINVAL;
588 if (end == start)
589 return 0;
097d5910 590 vma = find_vma(current->mm, start);
1da177e4
LT
591 if (!vma || vma->vm_start > start)
592 return -ENOMEM;
593
097d5910 594 prev = vma->vm_prev;
1da177e4
LT
595 if (start > vma->vm_start)
596 prev = vma;
597
598 for (nstart = start ; ; ) {
b0f205c2 599 vm_flags_t newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
1da177e4 600
1aab92ec 601 newflags |= flags;
1da177e4 602
1aab92ec 603 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
1da177e4
LT
604 tmp = vma->vm_end;
605 if (tmp > end)
606 tmp = end;
607 error = mlock_fixup(vma, &prev, nstart, tmp, newflags);
608 if (error)
609 break;
610 nstart = tmp;
611 if (nstart < prev->vm_end)
612 nstart = prev->vm_end;
613 if (nstart >= end)
614 break;
615
616 vma = prev->vm_next;
617 if (!vma || vma->vm_start != nstart) {
618 error = -ENOMEM;
619 break;
620 }
621 }
622 return error;
623}
624
0cf2f6f6
SG
625/*
626 * Go through vma areas and sum size of mlocked
627 * vma pages, as return value.
628 * Note deferred memory locking case(mlock2(,,MLOCK_ONFAULT)
629 * is also counted.
630 * Return value: previously mlocked page counts
631 */
632static int count_mm_mlocked_page_nr(struct mm_struct *mm,
633 unsigned long start, size_t len)
634{
635 struct vm_area_struct *vma;
636 int count = 0;
637
638 if (mm == NULL)
639 mm = current->mm;
640
641 vma = find_vma(mm, start);
642 if (vma == NULL)
643 vma = mm->mmap;
644
645 for (; vma ; vma = vma->vm_next) {
646 if (start >= vma->vm_end)
647 continue;
648 if (start + len <= vma->vm_start)
649 break;
650 if (vma->vm_flags & VM_LOCKED) {
651 if (start > vma->vm_start)
652 count -= (start - vma->vm_start);
653 if (start + len < vma->vm_end) {
654 count += start + len - vma->vm_start;
655 break;
656 }
657 count += vma->vm_end - vma->vm_start;
658 }
659 }
660
661 return count >> PAGE_SHIFT;
662}
663
dc0ef0df 664static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t flags)
1da177e4
LT
665{
666 unsigned long locked;
667 unsigned long lock_limit;
668 int error = -ENOMEM;
669
670 if (!can_do_mlock())
671 return -EPERM;
672
8fd9e488 673 len = PAGE_ALIGN(len + (offset_in_page(start)));
1da177e4
LT
674 start &= PAGE_MASK;
675
59e99e5b 676 lock_limit = rlimit(RLIMIT_MEMLOCK);
1da177e4 677 lock_limit >>= PAGE_SHIFT;
1f1cd705
DB
678 locked = len >> PAGE_SHIFT;
679
dc0ef0df
MH
680 if (down_write_killable(&current->mm->mmap_sem))
681 return -EINTR;
1f1cd705
DB
682
683 locked += current->mm->locked_vm;
0cf2f6f6
SG
684 if ((locked > lock_limit) && (!capable(CAP_IPC_LOCK))) {
685 /*
686 * It is possible that the regions requested intersect with
687 * previously mlocked areas, that part area in "mm->locked_vm"
688 * should not be counted to new mlock increment count. So check
689 * and adjust locked count if necessary.
690 */
691 locked -= count_mm_mlocked_page_nr(current->mm,
692 start, len);
693 }
1da177e4
LT
694
695 /* check against resource limits */
696 if ((locked <= lock_limit) || capable(CAP_IPC_LOCK))
1aab92ec 697 error = apply_vma_lock_flags(start, len, flags);
1f1cd705 698
1da177e4 699 up_write(&current->mm->mmap_sem);
c561259c
KS
700 if (error)
701 return error;
702
703 error = __mm_populate(start, len, 0);
704 if (error)
705 return __mlock_posix_error_return(error);
706 return 0;
1da177e4
LT
707}
708
1aab92ec
EM
709SYSCALL_DEFINE2(mlock, unsigned long, start, size_t, len)
710{
711 return do_mlock(start, len, VM_LOCKED);
712}
713
a8ca5d0e
EM
714SYSCALL_DEFINE3(mlock2, unsigned long, start, size_t, len, int, flags)
715{
b0f205c2
EM
716 vm_flags_t vm_flags = VM_LOCKED;
717
718 if (flags & ~MLOCK_ONFAULT)
a8ca5d0e
EM
719 return -EINVAL;
720
b0f205c2
EM
721 if (flags & MLOCK_ONFAULT)
722 vm_flags |= VM_LOCKONFAULT;
723
724 return do_mlock(start, len, vm_flags);
a8ca5d0e
EM
725}
726
6a6160a7 727SYSCALL_DEFINE2(munlock, unsigned long, start, size_t, len)
1da177e4
LT
728{
729 int ret;
730
8fd9e488 731 len = PAGE_ALIGN(len + (offset_in_page(start)));
1da177e4 732 start &= PAGE_MASK;
1f1cd705 733
dc0ef0df
MH
734 if (down_write_killable(&current->mm->mmap_sem))
735 return -EINTR;
1aab92ec 736 ret = apply_vma_lock_flags(start, len, 0);
1da177e4 737 up_write(&current->mm->mmap_sem);
1f1cd705 738
1da177e4
LT
739 return ret;
740}
741
b0f205c2
EM
742/*
743 * Take the MCL_* flags passed into mlockall (or 0 if called from munlockall)
744 * and translate into the appropriate modifications to mm->def_flags and/or the
745 * flags for all current VMAs.
746 *
747 * There are a couple of subtleties with this. If mlockall() is called multiple
748 * times with different flags, the values do not necessarily stack. If mlockall
749 * is called once including the MCL_FUTURE flag and then a second time without
750 * it, VM_LOCKED and VM_LOCKONFAULT will be cleared from mm->def_flags.
751 */
1aab92ec 752static int apply_mlockall_flags(int flags)
1da177e4
LT
753{
754 struct vm_area_struct * vma, * prev = NULL;
b0f205c2 755 vm_flags_t to_add = 0;
1da177e4 756
b0f205c2
EM
757 current->mm->def_flags &= VM_LOCKED_CLEAR_MASK;
758 if (flags & MCL_FUTURE) {
09a9f1d2 759 current->mm->def_flags |= VM_LOCKED;
1aab92ec 760
b0f205c2
EM
761 if (flags & MCL_ONFAULT)
762 current->mm->def_flags |= VM_LOCKONFAULT;
763
764 if (!(flags & MCL_CURRENT))
765 goto out;
766 }
767
768 if (flags & MCL_CURRENT) {
769 to_add |= VM_LOCKED;
770 if (flags & MCL_ONFAULT)
771 to_add |= VM_LOCKONFAULT;
772 }
1da177e4
LT
773
774 for (vma = current->mm->mmap; vma ; vma = prev->vm_next) {
ca16d140 775 vm_flags_t newflags;
1da177e4 776
b0f205c2
EM
777 newflags = vma->vm_flags & VM_LOCKED_CLEAR_MASK;
778 newflags |= to_add;
1da177e4
LT
779
780 /* Ignore errors */
781 mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
bde6c3aa 782 cond_resched_rcu_qs();
1da177e4
LT
783 }
784out:
785 return 0;
786}
787
3480b257 788SYSCALL_DEFINE1(mlockall, int, flags)
1da177e4
LT
789{
790 unsigned long lock_limit;
86d2adcc 791 int ret;
1da177e4 792
b0f205c2 793 if (!flags || (flags & ~(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT)))
86d2adcc 794 return -EINVAL;
1da177e4 795
1da177e4 796 if (!can_do_mlock())
86d2adcc 797 return -EPERM;
1da177e4 798
59e99e5b 799 lock_limit = rlimit(RLIMIT_MEMLOCK);
1da177e4
LT
800 lock_limit >>= PAGE_SHIFT;
801
dc0ef0df
MH
802 if (down_write_killable(&current->mm->mmap_sem))
803 return -EINTR;
1f1cd705 804
dc0ef0df 805 ret = -ENOMEM;
1da177e4
LT
806 if (!(flags & MCL_CURRENT) || (current->mm->total_vm <= lock_limit) ||
807 capable(CAP_IPC_LOCK))
1aab92ec 808 ret = apply_mlockall_flags(flags);
1da177e4 809 up_write(&current->mm->mmap_sem);
bebeb3d6
ML
810 if (!ret && (flags & MCL_CURRENT))
811 mm_populate(0, TASK_SIZE);
86d2adcc 812
1da177e4
LT
813 return ret;
814}
815
3480b257 816SYSCALL_DEFINE0(munlockall)
1da177e4
LT
817{
818 int ret;
819
dc0ef0df
MH
820 if (down_write_killable(&current->mm->mmap_sem))
821 return -EINTR;
1aab92ec 822 ret = apply_mlockall_flags(0);
1da177e4
LT
823 up_write(&current->mm->mmap_sem);
824 return ret;
825}
826
827/*
828 * Objects with different lifetime than processes (SHM_LOCK and SHM_HUGETLB
829 * shm segments) get accounted against the user_struct instead.
830 */
831static DEFINE_SPINLOCK(shmlock_user_lock);
832
833int user_shm_lock(size_t size, struct user_struct *user)
834{
835 unsigned long lock_limit, locked;
836 int allowed = 0;
837
838 locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
59e99e5b 839 lock_limit = rlimit(RLIMIT_MEMLOCK);
5ed44a40
HB
840 if (lock_limit == RLIM_INFINITY)
841 allowed = 1;
1da177e4
LT
842 lock_limit >>= PAGE_SHIFT;
843 spin_lock(&shmlock_user_lock);
5ed44a40
HB
844 if (!allowed &&
845 locked + user->locked_shm > lock_limit && !capable(CAP_IPC_LOCK))
1da177e4
LT
846 goto out;
847 get_uid(user);
848 user->locked_shm += locked;
849 allowed = 1;
850out:
851 spin_unlock(&shmlock_user_lock);
852 return allowed;
853}
854
855void user_shm_unlock(size_t size, struct user_struct *user)
856{
857 spin_lock(&shmlock_user_lock);
858 user->locked_shm -= (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
859 spin_unlock(&shmlock_user_lock);
860 free_uid(user);
861}