mm/mprotect.c

   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 <linux/pgtable.h>
  32 #include <linux/sched/sysctl.h>
  33 #include <linux/userfaultfd_k.h>
  34 #include <linux/memory-tiers.h>
  35 #include <asm/cacheflush.h>
  36 #include <asm/mmu_context.h>
  37 #include <asm/tlbflush.h>
  38 #include <asm/tlb.h>
  39
  40 #include "internal.h"
  41
  42 bool can_change_pte_writable(struct vm_area_struct *vma, unsigned long addr,
  43                              pte_t pte)
  44 {
  45         struct page *page;
  46
  47         if (WARN_ON_ONCE(!(vma->vm_flags & VM_WRITE)))
  48                 return false;
  49
  50         /* Don't touch entries that are not even readable. */
  51         if (pte_protnone(pte))
  52                 return false;
  53
  54         /* Do we need write faults for softdirty tracking? */
  55         if (vma_soft_dirty_enabled(vma) && !pte_soft_dirty(pte))
  56                 return false;
  57
  58         /* Do we need write faults for uffd-wp tracking? */
  59         if (userfaultfd_pte_wp(vma, pte))
  60                 return false;
  61
  62         if (!(vma->vm_flags & VM_SHARED)) {
  63                 /*
  64                  * Writable MAP_PRIVATE mapping: We can only special-case on
  65                  * exclusive anonymous pages, because we know that our
  66                  * write-fault handler similarly would map them writable without
  67                  * any additional checks while holding the PT lock.
  68                  */
  69                 page = vm_normal_page(vma, addr, pte);
  70                 return page && PageAnon(page) && PageAnonExclusive(page);
  71         }
  72
  73         /*
  74          * Writable MAP_SHARED mapping: "clean" might indicate that the FS still
  75          * needs a real write-fault for writenotify
  76          * (see vma_wants_writenotify()). If "dirty", the assumption is that the
  77          * FS was already notified and we can simply mark the PTE writable
  78          * just like the write-fault handler would do.
  79          */
  80         return pte_dirty(pte);
  81 }
  82
  83 static long change_pte_range(struct mmu_gather *tlb,
  84                 struct vm_area_struct *vma, pmd_t *pmd, unsigned long addr,
  85                 unsigned long end, pgprot_t newprot, unsigned long cp_flags)
  86 {
  87         pte_t *pte, oldpte;
  88         spinlock_t *ptl;
  89         long pages = 0;
  90         int target_node = NUMA_NO_NODE;
  91         bool prot_numa = cp_flags & MM_CP_PROT_NUMA;
  92         bool uffd_wp = cp_flags & MM_CP_UFFD_WP;
  93         bool uffd_wp_resolve = cp_flags & MM_CP_UFFD_WP_RESOLVE;
  94
  95         tlb_change_page_size(tlb, PAGE_SIZE);
  96         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
  97         if (!pte)
  98                 return -EAGAIN;
  99
 100         /* Get target node for single threaded private VMAs */
 101         if (prot_numa && !(vma->vm_flags & VM_SHARED) &&
 102             atomic_read(&vma->vm_mm->mm_users) == 1)
 103                 target_node = numa_node_id();
 104
 105         flush_tlb_batched_pending(vma->vm_mm);
 106         arch_enter_lazy_mmu_mode();
 107         do {
 108                 oldpte = ptep_get(pte);
 109                 if (pte_present(oldpte)) {
 110                         pte_t ptent;
 111
 112                         /*
 113                          * Avoid trapping faults against the zero or KSM
 114                          * pages. See similar comment in change_huge_pmd.
 115                          */
 116                         if (prot_numa) {
 117                                 struct folio *folio;
 118                                 int nid;
 119                                 bool toptier;
 120
 121                                 /* Avoid TLB flush if possible */
 122                                 if (pte_protnone(oldpte))
 123                                         continue;
 124
 125                                 folio = vm_normal_folio(vma, addr, oldpte);
 126                                 if (!folio || folio_is_zone_device(folio) ||
 127                                     folio_test_ksm(folio))
 128                                         continue;
 129
 130                                 /* Also skip shared copy-on-write pages */
 131                                 if (is_cow_mapping(vma->vm_flags) &&
 132                                     (folio_maybe_dma_pinned(folio) ||
 133                                      folio_likely_mapped_shared(folio)))
 134                                         continue;
 135
 136                                 /*
 137                                  * While migration can move some dirty pages,
 138                                  * it cannot move them all from MIGRATE_ASYNC
 139                                  * context.
 140                                  */
 141                                 if (folio_is_file_lru(folio) &&
 142                                     folio_test_dirty(folio))
 143                                         continue;
 144
 145                                 /*
 146                                  * Don't mess with PTEs if page is already on the node
 147                                  * a single-threaded process is running on.
 148                                  */
 149                                 nid = folio_nid(folio);
 150                                 if (target_node == nid)
 151                                         continue;
 152                                 toptier = node_is_toptier(nid);
 153
 154                                 /*
 155                                  * Skip scanning top tier node if normal numa
 156                                  * balancing is disabled
 157                                  */
 158                                 if (!(sysctl_numa_balancing_mode & NUMA_BALANCING_NORMAL) &&
 159                                     toptier)
 160                                         continue;
 161                                 if (sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING &&
 162                                     !toptier)
 163                                         folio_xchg_access_time(folio,
 164                                                 jiffies_to_msecs(jiffies));
 165                         }
 166
 167                         oldpte = ptep_modify_prot_start(vma, addr, pte);
 168                         ptent = pte_modify(oldpte, newprot);
 169
 170                         if (uffd_wp)
 171                                 ptent = pte_mkuffd_wp(ptent);
 172                         else if (uffd_wp_resolve)
 173                                 ptent = pte_clear_uffd_wp(ptent);
 174
 175                         /*
 176                          * In some writable, shared mappings, we might want
 177                          * to catch actual write access -- see
 178                          * vma_wants_writenotify().
 179                          *
 180                          * In all writable, private mappings, we have to
 181                          * properly handle COW.
 182                          *
 183                          * In both cases, we can sometimes still change PTEs
 184                          * writable and avoid the write-fault handler, for
 185                          * example, if a PTE is already dirty and no other
 186                          * COW or special handling is required.
 187                          */
 188                         if ((cp_flags & MM_CP_TRY_CHANGE_WRITABLE) &&
 189                             !pte_write(ptent) &&
 190                             can_change_pte_writable(vma, addr, ptent))
 191                                 ptent = pte_mkwrite(ptent, vma);
 192
 193                         ptep_modify_prot_commit(vma, addr, pte, oldpte, ptent);
 194                         if (pte_needs_flush(oldpte, ptent))
 195                                 tlb_flush_pte_range(tlb, addr, PAGE_SIZE);
 196                         pages++;
 197                 } else if (is_swap_pte(oldpte)) {
 198                         swp_entry_t entry = pte_to_swp_entry(oldpte);
 199                         pte_t newpte;
 200
 201                         if (is_writable_migration_entry(entry)) {
 202                                 struct folio *folio = pfn_swap_entry_folio(entry);
 203
 204                                 /*
 205                                  * A protection check is difficult so
 206                                  * just be safe and disable write
 207                                  */
 208                                 if (folio_test_anon(folio))
 209                                         entry = make_readable_exclusive_migration_entry(
 210                                                              swp_offset(entry));
 211                                 else
 212                                         entry = make_readable_migration_entry(swp_offset(entry));
 213                                 newpte = swp_entry_to_pte(entry);
 214                                 if (pte_swp_soft_dirty(oldpte))
 215                                         newpte = pte_swp_mksoft_dirty(newpte);
 216                         } else if (is_writable_device_private_entry(entry)) {
 217                                 /*
 218                                  * We do not preserve soft-dirtiness. See
 219                                  * copy_nonpresent_pte() for explanation.
 220                                  */
 221                                 entry = make_readable_device_private_entry(
 222                                                         swp_offset(entry));
 223                                 newpte = swp_entry_to_pte(entry);
 224                                 if (pte_swp_uffd_wp(oldpte))
 225                                         newpte = pte_swp_mkuffd_wp(newpte);
 226                         } else if (is_writable_device_exclusive_entry(entry)) {
 227                                 entry = make_readable_device_exclusive_entry(
 228                                                         swp_offset(entry));
 229                                 newpte = swp_entry_to_pte(entry);
 230                                 if (pte_swp_soft_dirty(oldpte))
 231                                         newpte = pte_swp_mksoft_dirty(newpte);
 232                                 if (pte_swp_uffd_wp(oldpte))
 233                                         newpte = pte_swp_mkuffd_wp(newpte);
 234                         } else if (is_pte_marker_entry(entry)) {
 235                                 /*
 236                                  * Ignore error swap entries unconditionally,
 237                                  * because any access should sigbus anyway.
 238                                  */
 239                                 if (is_poisoned_swp_entry(entry))
 240                                         continue;
 241                                 /*
 242                                  * If this is uffd-wp pte marker and we'd like
 243                                  * to unprotect it, drop it; the next page
 244                                  * fault will trigger without uffd trapping.
 245                                  */
 246                                 if (uffd_wp_resolve) {
 247                                         pte_clear(vma->vm_mm, addr, pte);
 248                                         pages++;
 249                                 }
 250                                 continue;
 251                         } else {
 252                                 newpte = oldpte;
 253                         }
 254
 255                         if (uffd_wp)
 256                                 newpte = pte_swp_mkuffd_wp(newpte);
 257                         else if (uffd_wp_resolve)
 258                                 newpte = pte_swp_clear_uffd_wp(newpte);
 259
 260                         if (!pte_same(oldpte, newpte)) {
 261                                 set_pte_at(vma->vm_mm, addr, pte, newpte);
 262                                 pages++;
 263                         }
 264                 } else {
 265                         /* It must be an none page, or what else?.. */
 266                         WARN_ON_ONCE(!pte_none(oldpte));
 267
 268                         /*
 269                          * Nobody plays with any none ptes besides
 270                          * userfaultfd when applying the protections.
 271                          */
 272                         if (likely(!uffd_wp))
 273                                 continue;
 274
 275                         if (userfaultfd_wp_use_markers(vma)) {
 276                                 /*
 277                                  * For file-backed mem, we need to be able to
 278                                  * wr-protect a none pte, because even if the
 279                                  * pte is none, the page/swap cache could
 280                                  * exist.  Doing that by install a marker.
 281                                  */
 282                                 set_pte_at(vma->vm_mm, addr, pte,
 283                                            make_pte_marker(PTE_MARKER_UFFD_WP));
 284                                 pages++;
 285                         }
 286                 }
 287         } while (pte++, addr += PAGE_SIZE, addr != end);
 288         arch_leave_lazy_mmu_mode();
 289         pte_unmap_unlock(pte - 1, ptl);
 290
 291         return pages;
 292 }
 293
 294 /*
 295  * Return true if we want to split THPs into PTE mappings in change
 296  * protection procedure, false otherwise.
 297  */
 298 static inline bool
 299 pgtable_split_needed(struct vm_area_struct *vma, unsigned long cp_flags)
 300 {
 301         /*
 302          * pte markers only resides in pte level, if we need pte markers,
 303          * we need to split.  We cannot wr-protect shmem thp because file
 304          * thp is handled differently when split by erasing the pmd so far.
 305          */
 306         return (cp_flags & MM_CP_UFFD_WP) && !vma_is_anonymous(vma);
 307 }
 308
 309 /*
 310  * Return true if we want to populate pgtables in change protection
 311  * procedure, false otherwise
 312  */
 313 static inline bool
 314 pgtable_populate_needed(struct vm_area_struct *vma, unsigned long cp_flags)
 315 {
 316         /* If not within ioctl(UFFDIO_WRITEPROTECT), then don't bother */
 317         if (!(cp_flags & MM_CP_UFFD_WP))
 318                 return false;
 319
 320         /* Populate if the userfaultfd mode requires pte markers */
 321         return userfaultfd_wp_use_markers(vma);
 322 }
 323
 324 /*
 325  * Populate the pgtable underneath for whatever reason if requested.
 326  * When {pte|pmd|...}_alloc() failed we treat it the same way as pgtable
 327  * allocation failures during page faults by kicking OOM and returning
 328  * error.
 329  */
 330 #define  change_pmd_prepare(vma, pmd, cp_flags)                         \
 331         ({                                                              \
 332                 long err = 0;                                           \
 333                 if (unlikely(pgtable_populate_needed(vma, cp_flags))) { \
 334                         if (pte_alloc(vma->vm_mm, pmd))                 \
 335                                 err = -ENOMEM;                          \
 336                 }                                                       \
 337                 err;                                                    \
 338         })
 339
 340 /*
 341  * This is the general pud/p4d/pgd version of change_pmd_prepare(). We need to
 342  * have separate change_pmd_prepare() because pte_alloc() returns 0 on success,
 343  * while {pmd|pud|p4d}_alloc() returns the valid pointer on success.
 344  */
 345 #define  change_prepare(vma, high, low, addr, cp_flags)                 \
 346           ({                                                            \
 347                 long err = 0;                                           \
 348                 if (unlikely(pgtable_populate_needed(vma, cp_flags))) { \
 349                         low##_t *p = low##_alloc(vma->vm_mm, high, addr); \
 350                         if (p == NULL)                                  \
 351                                 err = -ENOMEM;                          \
 352                 }                                                       \
 353                 err;                                                    \
 354         })
 355
 356 static inline long change_pmd_range(struct mmu_gather *tlb,
 357                 struct vm_area_struct *vma, pud_t *pud, unsigned long addr,
 358                 unsigned long end, pgprot_t newprot, unsigned long cp_flags)
 359 {
 360         pmd_t *pmd;
 361         unsigned long next;
 362         long pages = 0;
 363         unsigned long nr_huge_updates = 0;
 364         struct mmu_notifier_range range;
 365
 366         range.start = 0;
 367
 368         pmd = pmd_offset(pud, addr);
 369         do {
 370                 long ret;
 371                 pmd_t _pmd;
 372 again:
 373                 next = pmd_addr_end(addr, end);
 374
 375                 ret = change_pmd_prepare(vma, pmd, cp_flags);
 376                 if (ret) {
 377                         pages = ret;
 378                         break;
 379                 }
 380
 381                 if (pmd_none(*pmd))
 382                         goto next;
 383
 384                 /* invoke the mmu notifier if the pmd is populated */
 385                 if (!range.start) {
 386                         mmu_notifier_range_init(&range,
 387                                 MMU_NOTIFY_PROTECTION_VMA, 0,
 388                                 vma->vm_mm, addr, end);
 389                         mmu_notifier_invalidate_range_start(&range);
 390                 }
 391
 392                 _pmd = pmdp_get_lockless(pmd);
 393                 if (is_swap_pmd(_pmd) || pmd_trans_huge(_pmd) || pmd_devmap(_pmd)) {
 394                         if ((next - addr != HPAGE_PMD_SIZE) ||
 395                             pgtable_split_needed(vma, cp_flags)) {
 396                                 __split_huge_pmd(vma, pmd, addr, false, NULL);
 397                                 /*
 398                                  * For file-backed, the pmd could have been
 399                                  * cleared; make sure pmd populated if
 400                                  * necessary, then fall-through to pte level.
 401                                  */
 402                                 ret = change_pmd_prepare(vma, pmd, cp_flags);
 403                                 if (ret) {
 404                                         pages = ret;
 405                                         break;
 406                                 }
 407                         } else {
 408                                 ret = change_huge_pmd(tlb, vma, pmd,
 409                                                 addr, newprot, cp_flags);
 410                                 if (ret) {
 411                                         if (ret == HPAGE_PMD_NR) {
 412                                                 pages += HPAGE_PMD_NR;
 413                                                 nr_huge_updates++;
 414                                         }
 415
 416                                         /* huge pmd was handled */
 417                                         goto next;
 418                                 }
 419                         }
 420                         /* fall through, the trans huge pmd just split */
 421                 }
 422
 423                 ret = change_pte_range(tlb, vma, pmd, addr, next, newprot,
 424                                        cp_flags);
 425                 if (ret < 0)
 426                         goto again;
 427                 pages += ret;
 428 next:
 429                 cond_resched();
 430         } while (pmd++, addr = next, addr != end);
 431
 432         if (range.start)
 433                 mmu_notifier_invalidate_range_end(&range);
 434
 435         if (nr_huge_updates)
 436                 count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates);
 437         return pages;
 438 }
 439
 440 static inline long change_pud_range(struct mmu_gather *tlb,
 441                 struct vm_area_struct *vma, p4d_t *p4d, unsigned long addr,
 442                 unsigned long end, pgprot_t newprot, unsigned long cp_flags)
 443 {
 444         pud_t *pud;
 445         unsigned long next;
 446         long pages = 0, ret;
 447
 448         pud = pud_offset(p4d, addr);
 449         do {
 450                 next = pud_addr_end(addr, end);
 451                 ret = change_prepare(vma, pud, pmd, addr, cp_flags);
 452                 if (ret)
 453                         return ret;
 454                 if (pud_none_or_clear_bad(pud))
 455                         continue;
 456                 pages += change_pmd_range(tlb, vma, pud, addr, next, newprot,
 457                                           cp_flags);
 458         } while (pud++, addr = next, addr != end);
 459
 460         return pages;
 461 }
 462
 463 static inline long change_p4d_range(struct mmu_gather *tlb,
 464                 struct vm_area_struct *vma, pgd_t *pgd, unsigned long addr,
 465                 unsigned long end, pgprot_t newprot, unsigned long cp_flags)
 466 {
 467         p4d_t *p4d;
 468         unsigned long next;
 469         long pages = 0, ret;
 470
 471         p4d = p4d_offset(pgd, addr);
 472         do {
 473                 next = p4d_addr_end(addr, end);
 474                 ret = change_prepare(vma, p4d, pud, addr, cp_flags);
 475                 if (ret)
 476                         return ret;
 477                 if (p4d_none_or_clear_bad(p4d))
 478                         continue;
 479                 pages += change_pud_range(tlb, vma, p4d, addr, next, newprot,
 480                                           cp_flags);
 481         } while (p4d++, addr = next, addr != end);
 482
 483         return pages;
 484 }
 485
 486 static long change_protection_range(struct mmu_gather *tlb,
 487                 struct vm_area_struct *vma, unsigned long addr,
 488                 unsigned long end, pgprot_t newprot, unsigned long cp_flags)
 489 {
 490         struct mm_struct *mm = vma->vm_mm;
 491         pgd_t *pgd;
 492         unsigned long next;
 493         long pages = 0, ret;
 494
 495         BUG_ON(addr >= end);
 496         pgd = pgd_offset(mm, addr);
 497         tlb_start_vma(tlb, vma);
 498         do {
 499                 next = pgd_addr_end(addr, end);
 500                 ret = change_prepare(vma, pgd, p4d, addr, cp_flags);
 501                 if (ret) {
 502                         pages = ret;
 503                         break;
 504                 }
 505                 if (pgd_none_or_clear_bad(pgd))
 506                         continue;
 507                 pages += change_p4d_range(tlb, vma, pgd, addr, next, newprot,
 508                                           cp_flags);
 509         } while (pgd++, addr = next, addr != end);
 510
 511         tlb_end_vma(tlb, vma);
 512
 513         return pages;
 514 }
 515
 516 long change_protection(struct mmu_gather *tlb,
 517                        struct vm_area_struct *vma, unsigned long start,
 518                        unsigned long end, unsigned long cp_flags)
 519 {
 520         pgprot_t newprot = vma->vm_page_prot;
 521         long pages;
 522
 523         BUG_ON((cp_flags & MM_CP_UFFD_WP_ALL) == MM_CP_UFFD_WP_ALL);
 524
 525 #ifdef CONFIG_NUMA_BALANCING
 526         /*
 527          * Ordinary protection updates (mprotect, uffd-wp, softdirty tracking)
 528          * are expected to reflect their requirements via VMA flags such that
 529          * vma_set_page_prot() will adjust vma->vm_page_prot accordingly.
 530          */
 531         if (cp_flags & MM_CP_PROT_NUMA)
 532                 newprot = PAGE_NONE;
 533 #else
 534         WARN_ON_ONCE(cp_flags & MM_CP_PROT_NUMA);
 535 #endif
 536
 537         if (is_vm_hugetlb_page(vma))
 538                 pages = hugetlb_change_protection(vma, start, end, newprot,
 539                                                   cp_flags);
 540         else
 541                 pages = change_protection_range(tlb, vma, start, end, newprot,
 542                                                 cp_flags);
 543
 544         return pages;
 545 }
 546
 547 static int prot_none_pte_entry(pte_t *pte, unsigned long addr,
 548                                unsigned long next, struct mm_walk *walk)
 549 {
 550         return pfn_modify_allowed(pte_pfn(ptep_get(pte)),
 551                                   *(pgprot_t *)(walk->private)) ?
 552                 0 : -EACCES;
 553 }
 554
 555 static int prot_none_hugetlb_entry(pte_t *pte, unsigned long hmask,
 556                                    unsigned long addr, unsigned long next,
 557                                    struct mm_walk *walk)
 558 {
 559         return pfn_modify_allowed(pte_pfn(ptep_get(pte)),
 560                                   *(pgprot_t *)(walk->private)) ?
 561                 0 : -EACCES;
 562 }
 563
 564 static int prot_none_test(unsigned long addr, unsigned long next,
 565                           struct mm_walk *walk)
 566 {
 567         return 0;
 568 }
 569
 570 static const struct mm_walk_ops prot_none_walk_ops = {
 571         .pte_entry              = prot_none_pte_entry,
 572         .hugetlb_entry          = prot_none_hugetlb_entry,
 573         .test_walk              = prot_none_test,
 574         .walk_lock              = PGWALK_WRLOCK,
 575 };
 576
 577 int
 578 mprotect_fixup(struct vma_iterator *vmi, struct mmu_gather *tlb,
 579                struct vm_area_struct *vma, struct vm_area_struct **pprev,
 580                unsigned long start, unsigned long end, unsigned long newflags)
 581 {
 582         struct mm_struct *mm = vma->vm_mm;
 583         unsigned long oldflags = vma->vm_flags;
 584         long nrpages = (end - start) >> PAGE_SHIFT;
 585         unsigned int mm_cp_flags = 0;
 586         unsigned long charged = 0;
 587         int error;
 588
 589         if (newflags == oldflags) {
 590                 *pprev = vma;
 591                 return 0;
 592         }
 593
 594         /*
 595          * Do PROT_NONE PFN permission checks here when we can still
 596          * bail out without undoing a lot of state. This is a rather
 597          * uncommon case, so doesn't need to be very optimized.
 598          */
 599         if (arch_has_pfn_modify_check() &&
 600             (vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) &&
 601             (newflags & VM_ACCESS_FLAGS) == 0) {
 602                 pgprot_t new_pgprot = vm_get_page_prot(newflags);
 603
 604                 error = walk_page_range(current->mm, start, end,
 605                                 &prot_none_walk_ops, &new_pgprot);
 606                 if (error)
 607                         return error;
 608         }
 609
 610         /*
 611          * If we make a private mapping writable we increase our commit;
 612          * but (without finer accounting) cannot reduce our commit if we
 613          * make it unwritable again except in the anonymous case where no
 614          * anon_vma has yet to be assigned.
 615          *
 616          * hugetlb mapping were accounted for even if read-only so there is
 617          * no need to account for them here.
 618          */
 619         if (newflags & VM_WRITE) {
 620                 /* Check space limits when area turns into data. */
 621                 if (!may_expand_vm(mm, newflags, nrpages) &&
 622                                 may_expand_vm(mm, oldflags, nrpages))
 623                         return -ENOMEM;
 624                 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
 625                                                 VM_SHARED|VM_NORESERVE))) {
 626                         charged = nrpages;
 627                         if (security_vm_enough_memory_mm(mm, charged))
 628                                 return -ENOMEM;
 629                         newflags |= VM_ACCOUNT;
 630                 }
 631         } else if ((oldflags & VM_ACCOUNT) && vma_is_anonymous(vma) &&
 632                    !vma->anon_vma) {
 633                 newflags &= ~VM_ACCOUNT;
 634         }
 635
 636         vma = vma_modify_flags(vmi, *pprev, vma, start, end, newflags);
 637         if (IS_ERR(vma)) {
 638                 error = PTR_ERR(vma);
 639                 goto fail;
 640         }
 641
 642         *pprev = vma;
 643
 644         /*
 645          * vm_flags and vm_page_prot are protected by the mmap_lock
 646          * held in write mode.
 647          */
 648         vma_start_write(vma);
 649         vm_flags_reset(vma, newflags);
 650         if (vma_wants_manual_pte_write_upgrade(vma))
 651                 mm_cp_flags |= MM_CP_TRY_CHANGE_WRITABLE;
 652         vma_set_page_prot(vma);
 653
 654         change_protection(tlb, vma, start, end, mm_cp_flags);
 655
 656         if ((oldflags & VM_ACCOUNT) && !(newflags & VM_ACCOUNT))
 657                 vm_unacct_memory(nrpages);
 658
 659         /*
 660          * Private VM_LOCKED VMA becoming writable: trigger COW to avoid major
 661          * fault on access.
 662          */
 663         if ((oldflags & (VM_WRITE | VM_SHARED | VM_LOCKED)) == VM_LOCKED &&
 664                         (newflags & VM_WRITE)) {
 665                 populate_vma_page_range(vma, start, end, NULL);
 666         }
 667
 668         vm_stat_account(mm, oldflags, -nrpages);
 669         vm_stat_account(mm, newflags, nrpages);
 670         perf_event_mmap(vma);
 671         return 0;
 672
 673 fail:
 674         vm_unacct_memory(charged);
 675         return error;
 676 }
 677
 678 /*
 679  * pkey==-1 when doing a legacy mprotect()
 680  */
 681 static int do_mprotect_pkey(unsigned long start, size_t len,
 682                 unsigned long prot, int pkey)
 683 {
 684         unsigned long nstart, end, tmp, reqprot;
 685         struct vm_area_struct *vma, *prev;
 686         int error;
 687         const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
 688         const bool rier = (current->personality & READ_IMPLIES_EXEC) &&
 689                                 (prot & PROT_READ);
 690         struct mmu_gather tlb;
 691         struct vma_iterator vmi;
 692
 693         start = untagged_addr(start);
 694
 695         prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
 696         if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
 697                 return -EINVAL;
 698
 699         if (start & ~PAGE_MASK)
 700                 return -EINVAL;
 701         if (!len)
 702                 return 0;
 703         len = PAGE_ALIGN(len);
 704         end = start + len;
 705         if (end <= start)
 706                 return -ENOMEM;
 707         if (!arch_validate_prot(prot, start))
 708                 return -EINVAL;
 709
 710         reqprot = prot;
 711
 712         if (mmap_write_lock_killable(current->mm))
 713                 return -EINTR;
 714
 715         /*
 716          * If userspace did not allocate the pkey, do not let
 717          * them use it here.
 718          */
 719         error = -EINVAL;
 720         if ((pkey != -1) && !mm_pkey_is_allocated(current->mm, pkey))
 721                 goto out;
 722
 723         vma_iter_init(&vmi, current->mm, start);
 724         vma = vma_find(&vmi, end);
 725         error = -ENOMEM;
 726         if (!vma)
 727                 goto out;
 728
 729         if (unlikely(grows & PROT_GROWSDOWN)) {
 730                 if (vma->vm_start >= end)
 731                         goto out;
 732                 start = vma->vm_start;
 733                 error = -EINVAL;
 734                 if (!(vma->vm_flags & VM_GROWSDOWN))
 735                         goto out;
 736         } else {
 737                 if (vma->vm_start > start)
 738                         goto out;
 739                 if (unlikely(grows & PROT_GROWSUP)) {
 740                         end = vma->vm_end;
 741                         error = -EINVAL;
 742                         if (!(vma->vm_flags & VM_GROWSUP))
 743                                 goto out;
 744                 }
 745         }
 746
 747         prev = vma_prev(&vmi);
 748         if (start > vma->vm_start)
 749                 prev = vma;
 750
 751         tlb_gather_mmu(&tlb, current->mm);
 752         nstart = start;
 753         tmp = vma->vm_start;
 754         for_each_vma_range(vmi, vma, end) {
 755                 unsigned long mask_off_old_flags;
 756                 unsigned long newflags;
 757                 int new_vma_pkey;
 758
 759                 if (vma->vm_start != tmp) {
 760                         error = -ENOMEM;
 761                         break;
 762                 }
 763
 764                 /* Does the application expect PROT_READ to imply PROT_EXEC */
 765                 if (rier && (vma->vm_flags & VM_MAYEXEC))
 766                         prot |= PROT_EXEC;
 767
 768                 /*
 769                  * Each mprotect() call explicitly passes r/w/x permissions.
 770                  * If a permission is not passed to mprotect(), it must be
 771                  * cleared from the VMA.
 772                  */
 773                 mask_off_old_flags = VM_ACCESS_FLAGS | VM_FLAGS_CLEAR;
 774
 775                 new_vma_pkey = arch_override_mprotect_pkey(vma, prot, pkey);
 776                 newflags = calc_vm_prot_bits(prot, new_vma_pkey);
 777                 newflags |= (vma->vm_flags & ~mask_off_old_flags);
 778
 779                 /* newflags >> 4 shift VM_MAY% in place of VM_% */
 780                 if ((newflags & ~(newflags >> 4)) & VM_ACCESS_FLAGS) {
 781                         error = -EACCES;
 782                         break;
 783                 }
 784
 785                 if (map_deny_write_exec(vma, newflags)) {
 786                         error = -EACCES;
 787                         break;
 788                 }
 789
 790                 /* Allow architectures to sanity-check the new flags */
 791                 if (!arch_validate_flags(newflags)) {
 792                         error = -EINVAL;
 793                         break;
 794                 }
 795
 796                 error = security_file_mprotect(vma, reqprot, prot);
 797                 if (error)
 798                         break;
 799
 800                 tmp = vma->vm_end;
 801                 if (tmp > end)
 802                         tmp = end;
 803
 804                 if (vma->vm_ops && vma->vm_ops->mprotect) {
 805                         error = vma->vm_ops->mprotect(vma, nstart, tmp, newflags);
 806                         if (error)
 807                                 break;
 808                 }
 809
 810                 error = mprotect_fixup(&vmi, &tlb, vma, &prev, nstart, tmp, newflags);
 811                 if (error)
 812                         break;
 813
 814                 tmp = vma_iter_end(&vmi);
 815                 nstart = tmp;
 816                 prot = reqprot;
 817         }
 818         tlb_finish_mmu(&tlb);
 819
 820         if (!error && tmp < end)
 821                 error = -ENOMEM;
 822
 823 out:
 824         mmap_write_unlock(current->mm);
 825         return error;
 826 }
 827
 828 SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
 829                 unsigned long, prot)
 830 {
 831         return do_mprotect_pkey(start, len, prot, -1);
 832 }
 833
 834 #ifdef CONFIG_ARCH_HAS_PKEYS
 835
 836 SYSCALL_DEFINE4(pkey_mprotect, unsigned long, start, size_t, len,
 837                 unsigned long, prot, int, pkey)
 838 {
 839         return do_mprotect_pkey(start, len, prot, pkey);
 840 }
 841
 842 SYSCALL_DEFINE2(pkey_alloc, unsigned long, flags, unsigned long, init_val)
 843 {
 844         int pkey;
 845         int ret;
 846
 847         /* No flags supported yet. */
 848         if (flags)
 849                 return -EINVAL;
 850         /* check for unsupported init values */
 851         if (init_val & ~PKEY_ACCESS_MASK)
 852                 return -EINVAL;
 853
 854         mmap_write_lock(current->mm);
 855         pkey = mm_pkey_alloc(current->mm);
 856
 857         ret = -ENOSPC;
 858         if (pkey == -1)
 859                 goto out;
 860
 861         ret = arch_set_user_pkey_access(current, pkey, init_val);
 862         if (ret) {
 863                 mm_pkey_free(current->mm, pkey);
 864                 goto out;
 865         }
 866         ret = pkey;
 867 out:
 868         mmap_write_unlock(current->mm);
 869         return ret;
 870 }
 871
 872 SYSCALL_DEFINE1(pkey_free, int, pkey)
 873 {
 874         int ret;
 875
 876         mmap_write_lock(current->mm);
 877         ret = mm_pkey_free(current->mm, pkey);
 878         mmap_write_unlock(current->mm);
 879
 880         /*
 881          * We could provide warnings or errors if any VMA still
 882          * has the pkey set here.
 883          */
 884         return ret;
 885 }
 886
 887 #endif /* CONFIG_ARCH_HAS_PKEYS */