root/mm/mprotect.c

DEFINITIONS

1. change_pte_range
2. pmd_none_or_clear_bad_unless_trans_huge
3. change_pmd_range
4. change_pud_range
5. change_p4d_range
6. change_protection_range
7. change_protection
8. prot_none_pte_entry
9. prot_none_hugetlb_entry
10. prot_none_test
11. mprotect_fixup
12. do_mprotect_pkey
13. SYSCALL_DEFINE3
14. SYSCALL_DEFINE4
15. SYSCALL_DEFINE2
16. SYSCALL_DEFINE1

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  mm/mprotect.c
   4  *
   5  *  (C) Copyright 1994 Linus Torvalds
   6  *  (C) Copyright 2002 Christoph Hellwig
   7  *
   8  *  Address space accounting code       <alan@lxorguk.ukuu.org.uk>
   9  *  (C) Copyright 2002 Red Hat Inc, All Rights Reserved
  10  */
  11 
  12 #include <linux/pagewalk.h>
  13 #include <linux/hugetlb.h>
  14 #include <linux/shm.h>
  15 #include <linux/mman.h>
  16 #include <linux/fs.h>
  17 #include <linux/highmem.h>
  18 #include <linux/security.h>
  19 #include <linux/mempolicy.h>
  20 #include <linux/personality.h>
  21 #include <linux/syscalls.h>
  22 #include <linux/swap.h>
  23 #include <linux/swapops.h>
  24 #include <linux/mmu_notifier.h>
  25 #include <linux/migrate.h>
  26 #include <linux/perf_event.h>
  27 #include <linux/pkeys.h>
  28 #include <linux/ksm.h>
  29 #include <linux/uaccess.h>
  30 #include <linux/mm_inline.h>
  31 #include <asm/pgtable.h>
  32 #include <asm/cacheflush.h>
  33 #include <asm/mmu_context.h>
  34 #include <asm/tlbflush.h>
  35 
  36 #include "internal.h"
  37 
  38 static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
  39                 unsigned long addr, unsigned long end, pgprot_t newprot,
  40                 int dirty_accountable, int prot_numa)
  41 {
  42         pte_t *pte, oldpte;
  43         spinlock_t *ptl;
  44         unsigned long pages = 0;
  45         int target_node = NUMA_NO_NODE;
  46 
  47         /*
  48          * Can be called with only the mmap_sem for reading by
  49          * prot_numa so we must check the pmd isn't constantly
  50          * changing from under us from pmd_none to pmd_trans_huge
  51          * and/or the other way around.
  52          */
  53         if (pmd_trans_unstable(pmd))
  54                 return 0;
  55 
  56         /*
  57          * The pmd points to a regular pte so the pmd can't change
  58          * from under us even if the mmap_sem is only hold for
  59          * reading.
  60          */
  61         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
  62 
  63         /* Get target node for single threaded private VMAs */
  64         if (prot_numa && !(vma->vm_flags & VM_SHARED) &&
  65             atomic_read(&vma->vm_mm->mm_users) == 1)
  66                 target_node = numa_node_id();
  67 
  68         flush_tlb_batched_pending(vma->vm_mm);
  69         arch_enter_lazy_mmu_mode();
  70         do {
  71                 oldpte = *pte;
  72                 if (pte_present(oldpte)) {
  73                         pte_t ptent;
  74                         bool preserve_write = prot_numa && pte_write(oldpte);
  75 
  76                         /*
  77                          * Avoid trapping faults against the zero or KSM
  78                          * pages. See similar comment in change_huge_pmd.
  79                          */
  80                         if (prot_numa) {
  81                                 struct page *page;
  82 
  83                                 page = vm_normal_page(vma, addr, oldpte);
  84                                 if (!page || PageKsm(page))
  85                                         continue;
  86 
  87                                 /* Also skip shared copy-on-write pages */
  88                                 if (is_cow_mapping(vma->vm_flags) &&
  89                                     page_mapcount(page) != 1)
  90                                         continue;
  91 
  92                                 /*
  93                                  * While migration can move some dirty pages,
  94                                  * it cannot move them all from MIGRATE_ASYNC
  95                                  * context.
  96                                  */
  97                                 if (page_is_file_cache(page) && PageDirty(page))
  98                                         continue;
  99 
 100                                 /* Avoid TLB flush if possible */
 101                                 if (pte_protnone(oldpte))
 102                                         continue;
 103 
 104                                 /*
 105                                  * Don't mess with PTEs if page is already on the node
 106                                  * a single-threaded process is running on.
 107                                  */
 108                                 if (target_node == page_to_nid(page))
 109                                         continue;
 110                         }
 111 
 112                         oldpte = ptep_modify_prot_start(vma, addr, pte);
 113                         ptent = pte_modify(oldpte, newprot);
 114                         if (preserve_write)
 115                                 ptent = pte_mk_savedwrite(ptent);
 116 
 117                         /* Avoid taking write faults for known dirty pages */
 118                         if (dirty_accountable && pte_dirty(ptent) &&
 119                                         (pte_soft_dirty(ptent) ||
 120                                          !(vma->vm_flags & VM_SOFTDIRTY))) {
 121                                 ptent = pte_mkwrite(ptent);
 122                         }
 123                         ptep_modify_prot_commit(vma, addr, pte, oldpte, ptent);
 124                         pages++;
 125                 } else if (IS_ENABLED(CONFIG_MIGRATION)) {
 126                         swp_entry_t entry = pte_to_swp_entry(oldpte);
 127 
 128                         if (is_write_migration_entry(entry)) {
 129                                 pte_t newpte;
 130                                 /*
 131                                  * A protection check is difficult so
 132                                  * just be safe and disable write
 133                                  */
 134                                 make_migration_entry_read(&entry);
 135                                 newpte = swp_entry_to_pte(entry);
 136                                 if (pte_swp_soft_dirty(oldpte))
 137                                         newpte = pte_swp_mksoft_dirty(newpte);
 138                                 set_pte_at(vma->vm_mm, addr, pte, newpte);
 139 
 140                                 pages++;
 141                         }
 142 
 143                         if (is_write_device_private_entry(entry)) {
 144                                 pte_t newpte;
 145 
 146                                 /*
 147                                  * We do not preserve soft-dirtiness. See
 148                                  * copy_one_pte() for explanation.
 149                                  */
 150                                 make_device_private_entry_read(&entry);
 151                                 newpte = swp_entry_to_pte(entry);
 152                                 set_pte_at(vma->vm_mm, addr, pte, newpte);
 153 
 154                                 pages++;
 155                         }
 156                 }
 157         } while (pte++, addr += PAGE_SIZE, addr != end);
 158         arch_leave_lazy_mmu_mode();
 159         pte_unmap_unlock(pte - 1, ptl);
 160 
 161         return pages;
 162 }
 163 
 164 /*
 165  * Used when setting automatic NUMA hinting protection where it is
 166  * critical that a numa hinting PMD is not confused with a bad PMD.
 167  */
 168 static inline int pmd_none_or_clear_bad_unless_trans_huge(pmd_t *pmd)
 169 {
 170         pmd_t pmdval = pmd_read_atomic(pmd);
 171 
 172         /* See pmd_none_or_trans_huge_or_clear_bad for info on barrier */
 173 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 174         barrier();
 175 #endif
 176 
 177         if (pmd_none(pmdval))
 178                 return 1;
 179         if (pmd_trans_huge(pmdval))
 180                 return 0;
 181         if (unlikely(pmd_bad(pmdval))) {
 182                 pmd_clear_bad(pmd);
 183                 return 1;
 184         }
 185 
 186         return 0;
 187 }
 188 
 189 static inline unsigned long change_pmd_range(struct vm_area_struct *vma,
 190                 pud_t *pud, unsigned long addr, unsigned long end,
 191                 pgprot_t newprot, int dirty_accountable, int prot_numa)
 192 {
 193         pmd_t *pmd;
 194         unsigned long next;
 195         unsigned long pages = 0;
 196         unsigned long nr_huge_updates = 0;
 197         struct mmu_notifier_range range;
 198 
 199         range.start = 0;
 200 
 201         pmd = pmd_offset(pud, addr);
 202         do {
 203                 unsigned long this_pages;
 204 
 205                 next = pmd_addr_end(addr, end);
 206 
 207                 /*
 208                  * Automatic NUMA balancing walks the tables with mmap_sem
 209                  * held for read. It's possible a parallel update to occur
 210                  * between pmd_trans_huge() and a pmd_none_or_clear_bad()
 211                  * check leading to a false positive and clearing.
 212                  * Hence, it's necessary to atomically read the PMD value
 213                  * for all the checks.
 214                  */
 215                 if (!is_swap_pmd(*pmd) && !pmd_devmap(*pmd) &&
 216                      pmd_none_or_clear_bad_unless_trans_huge(pmd))
 217                         goto next;
 218 
 219                 /* invoke the mmu notifier if the pmd is populated */
 220                 if (!range.start) {
 221                         mmu_notifier_range_init(&range,
 222                                 MMU_NOTIFY_PROTECTION_VMA, 0,
 223                                 vma, vma->vm_mm, addr, end);
 224                         mmu_notifier_invalidate_range_start(&range);
 225                 }
 226 
 227                 if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) {
 228                         if (next - addr != HPAGE_PMD_SIZE) {
 229                                 __split_huge_pmd(vma, pmd, addr, false, NULL);
 230                         } else {
 231                                 int nr_ptes = change_huge_pmd(vma, pmd, addr,
 232                                                 newprot, prot_numa);
 233 
 234                                 if (nr_ptes) {
 235                                         if (nr_ptes == HPAGE_PMD_NR) {
 236                                                 pages += HPAGE_PMD_NR;
 237                                                 nr_huge_updates++;
 238                                         }
 239 
 240                                         /* huge pmd was handled */
 241                                         goto next;
 242                                 }
 243                         }
 244                         /* fall through, the trans huge pmd just split */
 245                 }
 246                 this_pages = change_pte_range(vma, pmd, addr, next, newprot,
 247                                  dirty_accountable, prot_numa);
 248                 pages += this_pages;
 249 next:
 250                 cond_resched();
 251         } while (pmd++, addr = next, addr != end);
 252 
 253         if (range.start)
 254                 mmu_notifier_invalidate_range_end(&range);
 255 
 256         if (nr_huge_updates)
 257                 count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates);
 258         return pages;
 259 }
 260 
 261 static inline unsigned long change_pud_range(struct vm_area_struct *vma,
 262                 p4d_t *p4d, unsigned long addr, unsigned long end,
 263                 pgprot_t newprot, int dirty_accountable, int prot_numa)
 264 {
 265         pud_t *pud;
 266         unsigned long next;
 267         unsigned long pages = 0;
 268 
 269         pud = pud_offset(p4d, addr);
 270         do {
 271                 next = pud_addr_end(addr, end);
 272                 if (pud_none_or_clear_bad(pud))
 273                         continue;
 274                 pages += change_pmd_range(vma, pud, addr, next, newprot,
 275                                  dirty_accountable, prot_numa);
 276         } while (pud++, addr = next, addr != end);
 277 
 278         return pages;
 279 }
 280 
 281 static inline unsigned long change_p4d_range(struct vm_area_struct *vma,
 282                 pgd_t *pgd, unsigned long addr, unsigned long end,
 283                 pgprot_t newprot, int dirty_accountable, int prot_numa)
 284 {
 285         p4d_t *p4d;
 286         unsigned long next;
 287         unsigned long pages = 0;
 288 
 289         p4d = p4d_offset(pgd, addr);
 290         do {
 291                 next = p4d_addr_end(addr, end);
 292                 if (p4d_none_or_clear_bad(p4d))
 293                         continue;
 294                 pages += change_pud_range(vma, p4d, addr, next, newprot,
 295                                  dirty_accountable, prot_numa);
 296         } while (p4d++, addr = next, addr != end);
 297 
 298         return pages;
 299 }
 300 
 301 static unsigned long change_protection_range(struct vm_area_struct *vma,
 302                 unsigned long addr, unsigned long end, pgprot_t newprot,
 303                 int dirty_accountable, int prot_numa)
 304 {
 305         struct mm_struct *mm = vma->vm_mm;
 306         pgd_t *pgd;
 307         unsigned long next;
 308         unsigned long start = addr;
 309         unsigned long pages = 0;
 310 
 311         BUG_ON(addr >= end);
 312         pgd = pgd_offset(mm, addr);
 313         flush_cache_range(vma, addr, end);
 314         inc_tlb_flush_pending(mm);
 315         do {
 316                 next = pgd_addr_end(addr, end);
 317                 if (pgd_none_or_clear_bad(pgd))
 318                         continue;
 319                 pages += change_p4d_range(vma, pgd, addr, next, newprot,
 320                                  dirty_accountable, prot_numa);
 321         } while (pgd++, addr = next, addr != end);
 322 
 323         /* Only flush the TLB if we actually modified any entries: */
 324         if (pages)
 325                 flush_tlb_range(vma, start, end);
 326         dec_tlb_flush_pending(mm);
 327 
 328         return pages;
 329 }
 330 
 331 unsigned long change_protection(struct vm_area_struct *vma, unsigned long start,
 332                        unsigned long end, pgprot_t newprot,
 333                        int dirty_accountable, int prot_numa)
 334 {
 335         unsigned long pages;
 336 
 337         if (is_vm_hugetlb_page(vma))
 338                 pages = hugetlb_change_protection(vma, start, end, newprot);
 339         else
 340                 pages = change_protection_range(vma, start, end, newprot, dirty_accountable, prot_numa);
 341 
 342         return pages;
 343 }
 344 
 345 static int prot_none_pte_entry(pte_t *pte, unsigned long addr,
 346                                unsigned long next, struct mm_walk *walk)
 347 {
 348         return pfn_modify_allowed(pte_pfn(*pte), *(pgprot_t *)(walk->private)) ?
 349                 0 : -EACCES;
 350 }
 351 
 352 static int prot_none_hugetlb_entry(pte_t *pte, unsigned long hmask,
 353                                    unsigned long addr, unsigned long next,
 354                                    struct mm_walk *walk)
 355 {
 356         return pfn_modify_allowed(pte_pfn(*pte), *(pgprot_t *)(walk->private)) ?
 357                 0 : -EACCES;
 358 }
 359 
 360 static int prot_none_test(unsigned long addr, unsigned long next,
 361                           struct mm_walk *walk)
 362 {
 363         return 0;
 364 }
 365 
 366 static const struct mm_walk_ops prot_none_walk_ops = {
 367         .pte_entry              = prot_none_pte_entry,
 368         .hugetlb_entry          = prot_none_hugetlb_entry,
 369         .test_walk              = prot_none_test,
 370 };
 371 
 372 int
 373 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
 374         unsigned long start, unsigned long end, unsigned long newflags)
 375 {
 376         struct mm_struct *mm = vma->vm_mm;
 377         unsigned long oldflags = vma->vm_flags;
 378         long nrpages = (end - start) >> PAGE_SHIFT;
 379         unsigned long charged = 0;
 380         pgoff_t pgoff;
 381         int error;
 382         int dirty_accountable = 0;
 383 
 384         if (newflags == oldflags) {
 385                 *pprev = vma;
 386                 return 0;
 387         }
 388 
 389         /*
 390          * Do PROT_NONE PFN permission checks here when we can still
 391          * bail out without undoing a lot of state. This is a rather
 392          * uncommon case, so doesn't need to be very optimized.
 393          */
 394         if (arch_has_pfn_modify_check() &&
 395             (vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) &&
 396             (newflags & (VM_READ|VM_WRITE|VM_EXEC)) == 0) {
 397                 pgprot_t new_pgprot = vm_get_page_prot(newflags);
 398 
 399                 error = walk_page_range(current->mm, start, end,
 400                                 &prot_none_walk_ops, &new_pgprot);
 401                 if (error)
 402                         return error;
 403         }
 404 
 405         /*
 406          * If we make a private mapping writable we increase our commit;
 407          * but (without finer accounting) cannot reduce our commit if we
 408          * make it unwritable again. hugetlb mapping were accounted for
 409          * even if read-only so there is no need to account for them here
 410          */
 411         if (newflags & VM_WRITE) {
 412                 /* Check space limits when area turns into data. */
 413                 if (!may_expand_vm(mm, newflags, nrpages) &&
 414                                 may_expand_vm(mm, oldflags, nrpages))
 415                         return -ENOMEM;
 416                 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
 417                                                 VM_SHARED|VM_NORESERVE))) {
 418                         charged = nrpages;
 419                         if (security_vm_enough_memory_mm(mm, charged))
 420                                 return -ENOMEM;
 421                         newflags |= VM_ACCOUNT;
 422                 }
 423         }
 424 
 425         /*
 426          * First try to merge with previous and/or next vma.
 427          */
 428         pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
 429         *pprev = vma_merge(mm, *pprev, start, end, newflags,
 430                            vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
 431                            vma->vm_userfaultfd_ctx);
 432         if (*pprev) {
 433                 vma = *pprev;
 434                 VM_WARN_ON((vma->vm_flags ^ newflags) & ~VM_SOFTDIRTY);
 435                 goto success;
 436         }
 437 
 438         *pprev = vma;
 439 
 440         if (start != vma->vm_start) {
 441                 error = split_vma(mm, vma, start, 1);
 442                 if (error)
 443                         goto fail;
 444         }
 445 
 446         if (end != vma->vm_end) {
 447                 error = split_vma(mm, vma, end, 0);
 448                 if (error)
 449                         goto fail;
 450         }
 451 
 452 success:
 453         /*
 454          * vm_flags and vm_page_prot are protected by the mmap_sem
 455          * held in write mode.
 456          */
 457         vma->vm_flags = newflags;
 458         dirty_accountable = vma_wants_writenotify(vma, vma->vm_page_prot);
 459         vma_set_page_prot(vma);
 460 
 461         change_protection(vma, start, end, vma->vm_page_prot,
 462                           dirty_accountable, 0);
 463 
 464         /*
 465          * Private VM_LOCKED VMA becoming writable: trigger COW to avoid major
 466          * fault on access.
 467          */
 468         if ((oldflags & (VM_WRITE | VM_SHARED | VM_LOCKED)) == VM_LOCKED &&
 469                         (newflags & VM_WRITE)) {
 470                 populate_vma_page_range(vma, start, end, NULL);
 471         }
 472 
 473         vm_stat_account(mm, oldflags, -nrpages);
 474         vm_stat_account(mm, newflags, nrpages);
 475         perf_event_mmap(vma);
 476         return 0;
 477 
 478 fail:
 479         vm_unacct_memory(charged);
 480         return error;
 481 }
 482 
 483 /*
 484  * pkey==-1 when doing a legacy mprotect()
 485  */
 486 static int do_mprotect_pkey(unsigned long start, size_t len,
 487                 unsigned long prot, int pkey)
 488 {
 489         unsigned long nstart, end, tmp, reqprot;
 490         struct vm_area_struct *vma, *prev;
 491         int error = -EINVAL;
 492         const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
 493         const bool rier = (current->personality & READ_IMPLIES_EXEC) &&
 494                                 (prot & PROT_READ);
 495 
 496         start = untagged_addr(start);
 497 
 498         prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
 499         if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
 500                 return -EINVAL;
 501 
 502         if (start & ~PAGE_MASK)
 503                 return -EINVAL;
 504         if (!len)
 505                 return 0;
 506         len = PAGE_ALIGN(len);
 507         end = start + len;
 508         if (end <= start)
 509                 return -ENOMEM;
 510         if (!arch_validate_prot(prot, start))
 511                 return -EINVAL;
 512 
 513         reqprot = prot;
 514 
 515         if (down_write_killable(&current->mm->mmap_sem))
 516                 return -EINTR;
 517 
 518         /*
 519          * If userspace did not allocate the pkey, do not let
 520          * them use it here.
 521          */
 522         error = -EINVAL;
 523         if ((pkey != -1) && !mm_pkey_is_allocated(current->mm, pkey))
 524                 goto out;
 525 
 526         vma = find_vma(current->mm, start);
 527         error = -ENOMEM;
 528         if (!vma)
 529                 goto out;
 530         prev = vma->vm_prev;
 531         if (unlikely(grows & PROT_GROWSDOWN)) {
 532                 if (vma->vm_start >= end)
 533                         goto out;
 534                 start = vma->vm_start;
 535                 error = -EINVAL;
 536                 if (!(vma->vm_flags & VM_GROWSDOWN))
 537                         goto out;
 538         } else {
 539                 if (vma->vm_start > start)
 540                         goto out;
 541                 if (unlikely(grows & PROT_GROWSUP)) {
 542                         end = vma->vm_end;
 543                         error = -EINVAL;
 544                         if (!(vma->vm_flags & VM_GROWSUP))
 545                                 goto out;
 546                 }
 547         }
 548         if (start > vma->vm_start)
 549                 prev = vma;
 550 
 551         for (nstart = start ; ; ) {
 552                 unsigned long mask_off_old_flags;
 553                 unsigned long newflags;
 554                 int new_vma_pkey;
 555 
 556                 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
 557 
 558                 /* Does the application expect PROT_READ to imply PROT_EXEC */
 559                 if (rier && (vma->vm_flags & VM_MAYEXEC))
 560                         prot |= PROT_EXEC;
 561 
 562                 /*
 563                  * Each mprotect() call explicitly passes r/w/x permissions.
 564                  * If a permission is not passed to mprotect(), it must be
 565                  * cleared from the VMA.
 566                  */
 567                 mask_off_old_flags = VM_READ | VM_WRITE | VM_EXEC |
 568                                         VM_FLAGS_CLEAR;
 569 
 570                 new_vma_pkey = arch_override_mprotect_pkey(vma, prot, pkey);
 571                 newflags = calc_vm_prot_bits(prot, new_vma_pkey);
 572                 newflags |= (vma->vm_flags & ~mask_off_old_flags);
 573 
 574                 /* newflags >> 4 shift VM_MAY% in place of VM_% */
 575                 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
 576                         error = -EACCES;
 577                         goto out;
 578                 }
 579 
 580                 error = security_file_mprotect(vma, reqprot, prot);
 581                 if (error)
 582                         goto out;
 583 
 584                 tmp = vma->vm_end;
 585                 if (tmp > end)
 586                         tmp = end;
 587                 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
 588                 if (error)
 589                         goto out;
 590                 nstart = tmp;
 591 
 592                 if (nstart < prev->vm_end)
 593                         nstart = prev->vm_end;
 594                 if (nstart >= end)
 595                         goto out;
 596 
 597                 vma = prev->vm_next;
 598                 if (!vma || vma->vm_start != nstart) {
 599                         error = -ENOMEM;
 600                         goto out;
 601                 }
 602                 prot = reqprot;
 603         }
 604 out:
 605         up_write(&current->mm->mmap_sem);
 606         return error;
 607 }
 608 
 609 SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
 610                 unsigned long, prot)
 611 {
 612         return do_mprotect_pkey(start, len, prot, -1);
 613 }
 614 
 615 #ifdef CONFIG_ARCH_HAS_PKEYS
 616 
 617 SYSCALL_DEFINE4(pkey_mprotect, unsigned long, start, size_t, len,
 618                 unsigned long, prot, int, pkey)
 619 {
 620         return do_mprotect_pkey(start, len, prot, pkey);
 621 }
 622 
 623 SYSCALL_DEFINE2(pkey_alloc, unsigned long, flags, unsigned long, init_val)
 624 {
 625         int pkey;
 626         int ret;
 627 
 628         /* No flags supported yet. */
 629         if (flags)
 630                 return -EINVAL;
 631         /* check for unsupported init values */
 632         if (init_val & ~PKEY_ACCESS_MASK)
 633                 return -EINVAL;
 634 
 635         down_write(&current->mm->mmap_sem);
 636         pkey = mm_pkey_alloc(current->mm);
 637 
 638         ret = -ENOSPC;
 639         if (pkey == -1)
 640                 goto out;
 641 
 642         ret = arch_set_user_pkey_access(current, pkey, init_val);
 643         if (ret) {
 644                 mm_pkey_free(current->mm, pkey);
 645                 goto out;
 646         }
 647         ret = pkey;
 648 out:
 649         up_write(&current->mm->mmap_sem);
 650         return ret;
 651 }
 652 
 653 SYSCALL_DEFINE1(pkey_free, int, pkey)
 654 {
 655         int ret;
 656 
 657         down_write(&current->mm->mmap_sem);
 658         ret = mm_pkey_free(current->mm, pkey);
 659         up_write(&current->mm->mmap_sem);
 660 
 661         /*
 662          * We could provie warnings or errors if any VMA still
 663          * has the pkey set here.
 664          */
 665         return ret;
 666 }
 667 
 668 #endif /* CONFIG_ARCH_HAS_PKEYS */