mm/mprotect.c

   1 /*
   2  *  mm/mprotect.c
   3  *
   4  *  (C) Copyright 1994 Linus Torvalds
   5  *  (C) Copyright 2002 Christoph Hellwig
   6  *
   7  *  Address space accounting code       <alan@lxorguk.ukuu.org.uk>
   8  *  (C) Copyright 2002 Red Hat Inc, All Rights Reserved
   9  */
  10
  11 #include <linux/mm.h>
  12 #include <linux/hugetlb.h>
  13 #include <linux/shm.h>
  14 #include <linux/mman.h>
  15 #include <linux/fs.h>
  16 #include <linux/highmem.h>
  17 #include <linux/security.h>
  18 #include <linux/mempolicy.h>
  19 #include <linux/personality.h>
  20 #include <linux/syscalls.h>
  21 #include <linux/swap.h>
  22 #include <linux/swapops.h>
  23 #include <linux/mmu_notifier.h>
  24 #include <linux/migrate.h>
  25 #include <linux/perf_event.h>
  26 #include <linux/ksm.h>
  27 #include <asm/uaccess.h>
  28 #include <asm/pgtable.h>
  29 #include <asm/cacheflush.h>
  30 #include <asm/tlbflush.h>
  31
  32 #ifndef pgprot_modify
  33 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
  34 {
  35         return newprot;
  36 }
  37 #endif
  38
  39 static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
  40                 unsigned long addr, unsigned long end, pgprot_t newprot,
  41                 int dirty_accountable, int prot_numa)
  42 {
  43         struct mm_struct *mm = vma->vm_mm;
  44         pte_t *pte, oldpte;
  45         spinlock_t *ptl;
  46         unsigned long pages = 0;
  47
  48         pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
  49         arch_enter_lazy_mmu_mode();
  50         do {
  51                 oldpte = *pte;
  52                 if (pte_present(oldpte)) {
  53                         pte_t ptent;
  54                         bool updated = false;
  55
  56                         if (!prot_numa) {
  57                                 ptent = ptep_modify_prot_start(mm, addr, pte);
  58                                 if (pte_numa(ptent))
  59                                         ptent = pte_mknonnuma(ptent);
  60                                 ptent = pte_modify(ptent, newprot);
  61                                 /*
  62                                  * Avoid taking write faults for pages we
  63                                  * know to be dirty.
  64                                  */
  65                                 if (dirty_accountable && pte_dirty(ptent))
  66                                         ptent = pte_mkwrite(ptent);
  67                                 ptep_modify_prot_commit(mm, addr, pte, ptent);
  68                                 updated = true;
  69                         } else {
  70                                 struct page *page;
  71
  72                                 page = vm_normal_page(vma, addr, oldpte);
  73                                 if (page && !PageKsm(page)) {
  74                                         if (!pte_numa(oldpte)) {
  75                                                 ptep_set_numa(mm, addr, pte);
  76                                                 updated = true;
  77                                         }
  78                                 }
  79                         }
  80                         if (updated)
  81                                 pages++;
  82                 } else if (IS_ENABLED(CONFIG_MIGRATION) && !pte_file(oldpte)) {
  83                         swp_entry_t entry = pte_to_swp_entry(oldpte);
  84
  85                         if (is_write_migration_entry(entry)) {
  86                                 pte_t newpte;
  87                                 /*
  88                                  * A protection check is difficult so
  89                                  * just be safe and disable write
  90                                  */
  91                                 make_migration_entry_read(&entry);
  92                                 newpte = swp_entry_to_pte(entry);
  93                                 if (pte_swp_soft_dirty(oldpte))
  94                                         newpte = pte_swp_mksoft_dirty(newpte);
  95                                 set_pte_at(mm, addr, pte, newpte);
  96
  97                                 pages++;
  98                         }
  99                 }
 100         } while (pte++, addr += PAGE_SIZE, addr != end);
 101         arch_leave_lazy_mmu_mode();
 102         pte_unmap_unlock(pte - 1, ptl);
 103
 104         return pages;
 105 }
 106
 107 static inline unsigned long change_pmd_range(struct vm_area_struct *vma,
 108                 pud_t *pud, unsigned long addr, unsigned long end,
 109                 pgprot_t newprot, int dirty_accountable, int prot_numa)
 110 {
 111         pmd_t *pmd;
 112         unsigned long next;
 113         unsigned long pages = 0;
 114         unsigned long nr_huge_updates = 0;
 115
 116         pmd = pmd_offset(pud, addr);
 117         do {
 118                 unsigned long this_pages;
 119
 120                 next = pmd_addr_end(addr, end);
 121                 if (pmd_trans_huge(*pmd)) {
 122                         if (next - addr != HPAGE_PMD_SIZE)
 123                                 split_huge_page_pmd(vma, addr, pmd);
 124                         else {
 125                                 int nr_ptes = change_huge_pmd(vma, pmd, addr,
 126                                                 newprot, prot_numa);
 127
 128                                 if (nr_ptes) {
 129                                         if (nr_ptes == HPAGE_PMD_NR) {
 130                                                 pages += HPAGE_PMD_NR;
 131                                                 nr_huge_updates++;
 132                                         }
 133                                         continue;
 134                                 }
 135                         }
 136                         /* fall through */
 137                 }
 138                 if (pmd_none_or_clear_bad(pmd))
 139                         continue;
 140                 this_pages = change_pte_range(vma, pmd, addr, next, newprot,
 141                                  dirty_accountable, prot_numa);
 142                 pages += this_pages;
 143         } while (pmd++, addr = next, addr != end);
 144
 145         if (nr_huge_updates)
 146                 count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates);
 147         return pages;
 148 }
 149
 150 static inline unsigned long change_pud_range(struct vm_area_struct *vma,
 151                 pgd_t *pgd, unsigned long addr, unsigned long end,
 152                 pgprot_t newprot, int dirty_accountable, int prot_numa)
 153 {
 154         pud_t *pud;
 155         unsigned long next;
 156         unsigned long pages = 0;
 157
 158         pud = pud_offset(pgd, addr);
 159         do {
 160                 next = pud_addr_end(addr, end);
 161                 if (pud_none_or_clear_bad(pud))
 162                         continue;
 163                 pages += change_pmd_range(vma, pud, addr, next, newprot,
 164                                  dirty_accountable, prot_numa);
 165         } while (pud++, addr = next, addr != end);
 166
 167         return pages;
 168 }
 169
 170 static unsigned long change_protection_range(struct vm_area_struct *vma,
 171                 unsigned long addr, unsigned long end, pgprot_t newprot,
 172                 int dirty_accountable, int prot_numa)
 173 {
 174         struct mm_struct *mm = vma->vm_mm;
 175         pgd_t *pgd;
 176         unsigned long next;
 177         unsigned long start = addr;
 178         unsigned long pages = 0;
 179
 180         BUG_ON(addr >= end);
 181         pgd = pgd_offset(mm, addr);
 182         flush_cache_range(vma, addr, end);
 183         set_tlb_flush_pending(mm);
 184         do {
 185                 next = pgd_addr_end(addr, end);
 186                 if (pgd_none_or_clear_bad(pgd))
 187                         continue;
 188                 pages += change_pud_range(vma, pgd, addr, next, newprot,
 189                                  dirty_accountable, prot_numa);
 190         } while (pgd++, addr = next, addr != end);
 191
 192         /* Only flush the TLB if we actually modified any entries: */
 193         if (pages)
 194                 flush_tlb_range(vma, start, end);
 195         clear_tlb_flush_pending(mm);
 196
 197         return pages;
 198 }
 199
 200 unsigned long change_protection(struct vm_area_struct *vma, unsigned long start,
 201                        unsigned long end, pgprot_t newprot,
 202                        int dirty_accountable, int prot_numa)
 203 {
 204         struct mm_struct *mm = vma->vm_mm;
 205         unsigned long pages;
 206
 207         mmu_notifier_invalidate_range_start(mm, start, end);
 208         if (is_vm_hugetlb_page(vma))
 209                 pages = hugetlb_change_protection(vma, start, end, newprot);
 210         else
 211                 pages = change_protection_range(vma, start, end, newprot, dirty_accountable, prot_numa);
 212         mmu_notifier_invalidate_range_end(mm, start, end);
 213
 214         return pages;
 215 }
 216
 217 int
 218 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
 219         unsigned long start, unsigned long end, unsigned long newflags)
 220 {
 221         struct mm_struct *mm = vma->vm_mm;
 222         unsigned long oldflags = vma->vm_flags;
 223         long nrpages = (end - start) >> PAGE_SHIFT;
 224         unsigned long charged = 0;
 225         pgoff_t pgoff;
 226         int error;
 227         int dirty_accountable = 0;
 228
 229         if (newflags == oldflags) {
 230                 *pprev = vma;
 231                 return 0;
 232         }
 233
 234         /*
 235          * If we make a private mapping writable we increase our commit;
 236          * but (without finer accounting) cannot reduce our commit if we
 237          * make it unwritable again. hugetlb mapping were accounted for
 238          * even if read-only so there is no need to account for them here
 239          */
 240         if (newflags & VM_WRITE) {
 241                 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
 242                                                 VM_SHARED|VM_NORESERVE))) {
 243                         charged = nrpages;
 244                         if (security_vm_enough_memory_mm(mm, charged))
 245                                 return -ENOMEM;
 246                         newflags |= VM_ACCOUNT;
 247                 }
 248         }
 249
 250         /*
 251          * First try to merge with previous and/or next vma.
 252          */
 253         pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
 254         *pprev = vma_merge(mm, *pprev, start, end, newflags,
 255                         vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
 256         if (*pprev) {
 257                 vma = *pprev;
 258                 goto success;
 259         }
 260
 261         *pprev = vma;
 262
 263         if (start != vma->vm_start) {
 264                 error = split_vma(mm, vma, start, 1);
 265                 if (error)
 266                         goto fail;
 267         }
 268
 269         if (end != vma->vm_end) {
 270                 error = split_vma(mm, vma, end, 0);
 271                 if (error)
 272                         goto fail;
 273         }
 274
 275 success:
 276         /*
 277          * vm_flags and vm_page_prot are protected by the mmap_sem
 278          * held in write mode.
 279          */
 280         vma->vm_flags = newflags;
 281         vma->vm_page_prot = pgprot_modify(vma->vm_page_prot,
 282                                           vm_get_page_prot(newflags));
 283
 284         if (vma_wants_writenotify(vma)) {
 285                 vma->vm_page_prot = vm_get_page_prot(newflags & ~VM_SHARED);
 286                 dirty_accountable = 1;
 287         }
 288
 289         change_protection(vma, start, end, vma->vm_page_prot,
 290                           dirty_accountable, 0);
 291
 292         vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
 293         vm_stat_account(mm, newflags, vma->vm_file, nrpages);
 294         perf_event_mmap(vma);
 295         return 0;
 296
 297 fail:
 298         vm_unacct_memory(charged);
 299         return error;
 300 }
 301
 302 SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
 303                 unsigned long, prot)
 304 {
 305         unsigned long vm_flags, nstart, end, tmp, reqprot;
 306         struct vm_area_struct *vma, *prev;
 307         int error = -EINVAL;
 308         const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
 309         prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
 310         if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
 311                 return -EINVAL;
 312
 313         if (start & ~PAGE_MASK)
 314                 return -EINVAL;
 315         if (!len)
 316                 return 0;
 317         len = PAGE_ALIGN(len);
 318         end = start + len;
 319         if (end <= start)
 320                 return -ENOMEM;
 321         if (!arch_validate_prot(prot))
 322                 return -EINVAL;
 323
 324         reqprot = prot;
 325         /*
 326          * Does the application expect PROT_READ to imply PROT_EXEC:
 327          */
 328         if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
 329                 prot |= PROT_EXEC;
 330
 331         vm_flags = calc_vm_prot_bits(prot);
 332
 333         down_write(&current->mm->mmap_sem);
 334
 335         vma = find_vma(current->mm, start);
 336         error = -ENOMEM;
 337         if (!vma)
 338                 goto out;
 339         prev = vma->vm_prev;
 340         if (unlikely(grows & PROT_GROWSDOWN)) {
 341                 if (vma->vm_start >= end)
 342                         goto out;
 343                 start = vma->vm_start;
 344                 error = -EINVAL;
 345                 if (!(vma->vm_flags & VM_GROWSDOWN))
 346                         goto out;
 347         } else {
 348                 if (vma->vm_start > start)
 349                         goto out;
 350                 if (unlikely(grows & PROT_GROWSUP)) {
 351                         end = vma->vm_end;
 352                         error = -EINVAL;
 353                         if (!(vma->vm_flags & VM_GROWSUP))
 354                                 goto out;
 355                 }
 356         }
 357         if (start > vma->vm_start)
 358                 prev = vma;
 359
 360         for (nstart = start ; ; ) {
 361                 unsigned long newflags;
 362
 363                 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
 364
 365                 newflags = vm_flags;
 366                 newflags |= (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
 367
 368                 /* newflags >> 4 shift VM_MAY% in place of VM_% */
 369                 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
 370                         error = -EACCES;
 371                         goto out;
 372                 }
 373
 374                 error = security_file_mprotect(vma, reqprot, prot);
 375                 if (error)
 376                         goto out;
 377
 378                 tmp = vma->vm_end;
 379                 if (tmp > end)
 380                         tmp = end;
 381                 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
 382                 if (error)
 383                         goto out;
 384                 nstart = tmp;
 385
 386                 if (nstart < prev->vm_end)
 387                         nstart = prev->vm_end;
 388                 if (nstart >= end)
 389                         goto out;
 390
 391                 vma = prev->vm_next;
 392                 if (!vma || vma->vm_start != nstart) {
 393                         error = -ENOMEM;
 394                         goto out;
 395                 }
 396         }
 397 out:
 398         up_write(&current->mm->mmap_sem);
 399         return error;
 400 }