mm/mlock.c

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