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 <asm/uaccess.h>
  27 #include <asm/pgtable.h>
  28 #include <asm/pgalloc.h>
  29 #include <asm/cacheflush.h>
  30 #include <asm/tlbflush.h>
  31
  32 #ifndef arch_remove_exec_range
  33 #define arch_remove_exec_range(mm, limit)      do { ; } while (0)
  34 #endif
  35
  36 #ifndef pgprot_modify
  37 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
  38 {
  39         return newprot;
  40 }
  41 #endif
  42
  43 static void change_pte_range(struct mm_struct *mm, pmd_t *pmd,
  44                 unsigned long addr, unsigned long end, pgprot_t newprot,
  45                 int dirty_accountable)
  46 {
  47         pte_t *pte, oldpte;
  48         spinlock_t *ptl;
  49
  50         pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
  51         arch_enter_lazy_mmu_mode();
  52         do {
  53                 oldpte = *pte;
  54                 if (pte_present(oldpte)) {
  55                         pte_t ptent;
  56
  57                         ptent = ptep_modify_prot_start(mm, addr, pte);
  58                         ptent = pte_modify(ptent, newprot);
  59
  60                         /*
  61                          * Avoid taking write faults for pages we know to be
  62                          * dirty.
  63                          */
  64                         if (dirty_accountable && pte_dirty(ptent))
  65                                 ptent = pte_mkwrite(ptent);
  66
  67                         ptep_modify_prot_commit(mm, addr, pte, ptent);
  68                 } else if (PAGE_MIGRATION && !pte_file(oldpte)) {
  69                         swp_entry_t entry = pte_to_swp_entry(oldpte);
  70
  71                         if (is_write_migration_entry(entry)) {
  72                                 /*
  73                                  * A protection check is difficult so
  74                                  * just be safe and disable write
  75                                  */
  76                                 make_migration_entry_read(&entry);
  77                                 set_pte_at(mm, addr, pte,
  78                                         swp_entry_to_pte(entry));
  79                         }
  80                 }
  81         } while (pte++, addr += PAGE_SIZE, addr != end);
  82         arch_leave_lazy_mmu_mode();
  83         pte_unmap_unlock(pte - 1, ptl);
  84 }
  85
  86 static inline void change_pmd_range(struct vm_area_struct *vma, pud_t *pud,
  87                 unsigned long addr, unsigned long end, pgprot_t newprot,
  88                 int dirty_accountable)
  89 {
  90         pmd_t *pmd;
  91         unsigned long next;
  92
  93         pmd = pmd_offset(pud, addr);
  94         do {
  95                 next = pmd_addr_end(addr, end);
  96                 if (pmd_trans_huge(*pmd)) {
  97                         if (next - addr != HPAGE_PMD_SIZE)
  98                                 split_huge_page_pmd(vma->vm_mm, pmd);
  99                         else if (change_huge_pmd(vma, pmd, addr, newprot))
 100                                 continue;
 101                         /* fall through */
 102                 }
 103                 if (pmd_none_or_clear_bad(pmd))
 104                         continue;
 105                 change_pte_range(vma->vm_mm, pmd, addr, next, newprot,
 106                                  dirty_accountable);
 107         } while (pmd++, addr = next, addr != end);
 108 }
 109
 110 static inline void change_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
 111                 unsigned long addr, unsigned long end, pgprot_t newprot,
 112                 int dirty_accountable)
 113 {
 114         pud_t *pud;
 115         unsigned long next;
 116
 117         pud = pud_offset(pgd, addr);
 118         do {
 119                 next = pud_addr_end(addr, end);
 120                 if (pud_none_or_clear_bad(pud))
 121                         continue;
 122                 change_pmd_range(vma, pud, addr, next, newprot,
 123                                  dirty_accountable);
 124         } while (pud++, addr = next, addr != end);
 125 }
 126
 127 static void change_protection(struct vm_area_struct *vma,
 128                 unsigned long addr, unsigned long end, pgprot_t newprot,
 129                 int dirty_accountable)
 130 {
 131         struct mm_struct *mm = vma->vm_mm;
 132         pgd_t *pgd;
 133         unsigned long next;
 134         unsigned long start = addr;
 135
 136         BUG_ON(addr >= end);
 137         pgd = pgd_offset(mm, addr);
 138         flush_cache_range(vma, addr, end);
 139         do {
 140                 next = pgd_addr_end(addr, end);
 141                 if (pgd_none_or_clear_bad(pgd))
 142                         continue;
 143                 change_pud_range(vma, pgd, addr, next, newprot,
 144                                  dirty_accountable);
 145         } while (pgd++, addr = next, addr != end);
 146         flush_tlb_range(vma, start, end);
 147 }
 148
 149 int
 150 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
 151         unsigned long start, unsigned long end, unsigned long newflags)
 152 {
 153         struct mm_struct *mm = vma->vm_mm;
 154         unsigned long oldflags = vma->vm_flags;
 155         long nrpages = (end - start) >> PAGE_SHIFT;
 156         unsigned long charged = 0, old_end = vma->vm_end;
 157         pgoff_t pgoff;
 158         int error;
 159         int dirty_accountable = 0;
 160
 161         if (newflags == oldflags) {
 162                 *pprev = vma;
 163                 return 0;
 164         }
 165
 166         /*
 167          * If we make a private mapping writable we increase our commit;
 168          * but (without finer accounting) cannot reduce our commit if we
 169          * make it unwritable again. hugetlb mapping were accounted for
 170          * even if read-only so there is no need to account for them here
 171          */
 172         if (newflags & VM_WRITE) {
 173                 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
 174                                                 VM_SHARED|VM_NORESERVE))) {
 175                         charged = nrpages;
 176                         if (security_vm_enough_memory(charged))
 177                                 return -ENOMEM;
 178                         newflags |= VM_ACCOUNT;
 179                 }
 180         }
 181
 182         /*
 183          * First try to merge with previous and/or next vma.
 184          */
 185         pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
 186         *pprev = vma_merge(mm, *pprev, start, end, newflags,
 187                         vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
 188         if (*pprev) {
 189                 vma = *pprev;
 190                 goto success;
 191         }
 192
 193         *pprev = vma;
 194
 195         if (start != vma->vm_start) {
 196                 error = split_vma(mm, vma, start, 1);
 197                 if (error)
 198                         goto fail;
 199         }
 200
 201         if (end != vma->vm_end) {
 202                 error = split_vma(mm, vma, end, 0);
 203                 if (error)
 204                         goto fail;
 205         }
 206
 207 success:
 208         /*
 209          * vm_flags and vm_page_prot are protected by the mmap_sem
 210          * held in write mode.
 211          */
 212         vma->vm_flags = newflags;
 213         vma->vm_page_prot = pgprot_modify(vma->vm_page_prot,
 214                                           vm_get_page_prot(newflags));
 215
 216         if (vma_wants_writenotify(vma)) {
 217                 vma->vm_page_prot = vm_get_page_prot(newflags & ~VM_SHARED);
 218                 dirty_accountable = 1;
 219         }
 220
 221         if (oldflags & VM_EXEC)
 222                 arch_remove_exec_range(current->mm, old_end);
 223
 224         mmu_notifier_invalidate_range_start(mm, start, end);
 225         if (is_vm_hugetlb_page(vma))
 226                 hugetlb_change_protection(vma, start, end, vma->vm_page_prot);
 227         else
 228                 change_protection(vma, start, end, vma->vm_page_prot, dirty_accountable);
 229         mmu_notifier_invalidate_range_end(mm, start, end);
 230         vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
 231         vm_stat_account(mm, newflags, vma->vm_file, nrpages);
 232         perf_event_mmap(vma);
 233         return 0;
 234
 235 fail:
 236         vm_unacct_memory(charged);
 237         return error;
 238 }
 239
 240 SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
 241                 unsigned long, prot)
 242 {
 243         unsigned long vm_flags, nstart, end, tmp, reqprot;
 244         struct vm_area_struct *vma, *prev;
 245         int error = -EINVAL;
 246         const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
 247         prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
 248         if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
 249                 return -EINVAL;
 250
 251         if (start & ~PAGE_MASK)
 252                 return -EINVAL;
 253         if (!len)
 254                 return 0;
 255         len = PAGE_ALIGN(len);
 256         end = start + len;
 257         if (end <= start)
 258                 return -ENOMEM;
 259         if (!arch_validate_prot(prot))
 260                 return -EINVAL;
 261
 262         reqprot = prot;
 263         /*
 264          * Does the application expect PROT_READ to imply PROT_EXEC:
 265          */
 266         if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
 267                 prot |= PROT_EXEC;
 268
 269         vm_flags = calc_vm_prot_bits(prot);
 270
 271         down_write(&current->mm->mmap_sem);
 272
 273         vma = find_vma_prev(current->mm, start, &prev);
 274         error = -ENOMEM;
 275         if (!vma)
 276                 goto out;
 277         if (unlikely(grows & PROT_GROWSDOWN)) {
 278                 if (vma->vm_start >= end)
 279                         goto out;
 280                 start = vma->vm_start;
 281                 error = -EINVAL;
 282                 if (!(vma->vm_flags & VM_GROWSDOWN))
 283                         goto out;
 284         }
 285         else {
 286                 if (vma->vm_start > start)
 287                         goto out;
 288                 if (unlikely(grows & PROT_GROWSUP)) {
 289                         end = vma->vm_end;
 290                         error = -EINVAL;
 291                         if (!(vma->vm_flags & VM_GROWSUP))
 292                                 goto out;
 293                 }
 294         }
 295         if (start > vma->vm_start)
 296                 prev = vma;
 297
 298         for (nstart = start ; ; ) {
 299                 unsigned long newflags;
 300
 301                 /* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
 302
 303                 newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
 304
 305                 /* newflags >> 4 shift VM_MAY% in place of VM_% */
 306                 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
 307                         error = -EACCES;
 308                         goto out;
 309                 }
 310
 311                 error = security_file_mprotect(vma, reqprot, prot);
 312                 if (error)
 313                         goto out;
 314
 315                 tmp = vma->vm_end;
 316                 if (tmp > end)
 317                         tmp = end;
 318                 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
 319                 if (error)
 320                         goto out;
 321                 nstart = tmp;
 322
 323                 if (nstart < prev->vm_end)
 324                         nstart = prev->vm_end;
 325                 if (nstart >= end)
 326                         goto out;
 327
 328                 vma = prev->vm_next;
 329                 if (!vma || vma->vm_start != nstart) {
 330                         error = -ENOMEM;
 331                         goto out;
 332                 }
 333         }
 334 out:
 335         up_write(&current->mm->mmap_sem);
 336         return error;
 337 }