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@redhat.com>
   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/slab.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
  23 #include <asm/uaccess.h>
  24 #include <asm/pgtable.h>
  25 #include <asm/pgalloc.h>
  26 #include <asm/cacheflush.h>
  27 #include <asm/tlbflush.h>
  28
  29 static void change_pte_range(struct mm_struct *mm, pmd_t *pmd,
  30                 unsigned long addr, unsigned long end, pgprot_t newprot)
  31 {
  32         pte_t *pte;
  33         spinlock_t *ptl;
  34
  35         pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
  36         do {
  37                 if (pte_present(*pte)) {
  38                         pte_t ptent;
  39
  40                         /* Avoid an SMP race with hardware updated dirty/clean
  41                          * bits by wiping the pte and then setting the new pte
  42                          * into place.
  43                          */
  44                         ptent = pte_modify(ptep_get_and_clear(mm, addr, pte), newprot);
  45                         set_pte_at(mm, addr, pte, ptent);
  46                         lazy_mmu_prot_update(ptent);
  47                 }
  48         } while (pte++, addr += PAGE_SIZE, addr != end);
  49         pte_unmap_unlock(pte - 1, ptl);
  50 }
  51
  52 static inline void change_pmd_range(struct mm_struct *mm, pud_t *pud,
  53                 unsigned long addr, unsigned long end, pgprot_t newprot)
  54 {
  55         pmd_t *pmd;
  56         unsigned long next;
  57
  58         pmd = pmd_offset(pud, addr);
  59         do {
  60                 next = pmd_addr_end(addr, end);
  61                 if (pmd_none_or_clear_bad(pmd))
  62                         continue;
  63                 change_pte_range(mm, pmd, addr, next, newprot);
  64         } while (pmd++, addr = next, addr != end);
  65 }
  66
  67 static inline void change_pud_range(struct mm_struct *mm, pgd_t *pgd,
  68                 unsigned long addr, unsigned long end, pgprot_t newprot)
  69 {
  70         pud_t *pud;
  71         unsigned long next;
  72
  73         pud = pud_offset(pgd, addr);
  74         do {
  75                 next = pud_addr_end(addr, end);
  76                 if (pud_none_or_clear_bad(pud))
  77                         continue;
  78                 change_pmd_range(mm, pud, addr, next, newprot);
  79         } while (pud++, addr = next, addr != end);
  80 }
  81
  82 static void change_protection(struct vm_area_struct *vma,
  83                 unsigned long addr, unsigned long end, pgprot_t newprot)
  84 {
  85         struct mm_struct *mm = vma->vm_mm;
  86         pgd_t *pgd;
  87         unsigned long next;
  88         unsigned long start = addr;
  89
  90         BUG_ON(addr >= end);
  91         pgd = pgd_offset(mm, addr);
  92         flush_cache_range(vma, addr, end);
  93         do {
  94                 next = pgd_addr_end(addr, end);
  95                 if (pgd_none_or_clear_bad(pgd))
  96                         continue;
  97                 change_pud_range(mm, pgd, addr, next, newprot);
  98         } while (pgd++, addr = next, addr != end);
  99         flush_tlb_range(vma, start, end);
 100 }
 101
 102 static int
 103 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
 104         unsigned long start, unsigned long end, unsigned long newflags)
 105 {
 106         struct mm_struct *mm = vma->vm_mm;
 107         unsigned long oldflags = vma->vm_flags;
 108         long nrpages = (end - start) >> PAGE_SHIFT;
 109         unsigned long charged = 0, old_end = vma->vm_end;
 110         pgprot_t newprot;
 111         pgoff_t pgoff;
 112         int error;
 113
 114         if (newflags == oldflags) {
 115                 *pprev = vma;
 116                 return 0;
 117         }
 118
 119         /*
 120          * If we make a private mapping writable we increase our commit;
 121          * but (without finer accounting) cannot reduce our commit if we
 122          * make it unwritable again.
 123          *
 124          * FIXME? We haven't defined a VM_NORESERVE flag, so mprotecting
 125          * a MAP_NORESERVE private mapping to writable will now reserve.
 126          */
 127         if (newflags & VM_WRITE) {
 128                 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_SHARED))) {
 129                         charged = nrpages;
 130                         if (security_vm_enough_memory(charged))
 131                                 return -ENOMEM;
 132                         newflags |= VM_ACCOUNT;
 133                 }
 134         }
 135
 136         newprot = protection_map[newflags & 0xf];
 137
 138         /*
 139          * First try to merge with previous and/or next vma.
 140          */
 141         pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
 142         *pprev = vma_merge(mm, *pprev, start, end, newflags,
 143                         vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
 144         if (*pprev) {
 145                 vma = *pprev;
 146                 goto success;
 147         }
 148
 149         *pprev = vma;
 150
 151         if (start != vma->vm_start) {
 152                 error = split_vma(mm, vma, start, 1);
 153                 if (error)
 154                         goto fail;
 155         }
 156
 157         if (end != vma->vm_end) {
 158                 error = split_vma(mm, vma, end, 0);
 159                 if (error)
 160                         goto fail;
 161         }
 162
 163 success:
 164         /*
 165          * vm_flags and vm_page_prot are protected by the mmap_sem
 166          * held in write mode.
 167          */
 168         vma->vm_flags = newflags;
 169         vma->vm_page_prot = newprot;
 170         if (oldflags & VM_EXEC)
 171                 arch_remove_exec_range(current->mm, old_end);
 172         if (is_vm_hugetlb_page(vma))
 173                 hugetlb_change_protection(vma, start, end, newprot);
 174         else
 175                 change_protection(vma, start, end, newprot);
 176         vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
 177         vm_stat_account(mm, newflags, vma->vm_file, nrpages);
 178         return 0;
 179
 180 fail:
 181         vm_unacct_memory(charged);
 182         return error;
 183 }
 184
 185 asmlinkage long
 186 sys_mprotect(unsigned long start, size_t len, unsigned long prot)
 187 {
 188         unsigned long vm_flags, nstart, end, tmp, reqprot;
 189         struct vm_area_struct *vma, *prev;
 190         int error = -EINVAL;
 191         const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
 192         prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
 193         if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
 194                 return -EINVAL;
 195
 196         if (start & ~PAGE_MASK)
 197                 return -EINVAL;
 198         if (!len)
 199                 return 0;
 200         len = PAGE_ALIGN(len);
 201         end = start + len;
 202         if (end <= start)
 203                 return -ENOMEM;
 204         if (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC | PROT_SEM))
 205                 return -EINVAL;
 206
 207         reqprot = prot;
 208         /*
 209          * Does the application expect PROT_READ to imply PROT_EXEC:
 210          */
 211         if (unlikely((prot & PROT_READ) &&
 212                         (current->personality & READ_IMPLIES_EXEC)))
 213                 prot |= PROT_EXEC;
 214
 215         vm_flags = calc_vm_prot_bits(prot);
 216
 217         down_write(&current->mm->mmap_sem);
 218
 219         vma = find_vma_prev(current->mm, start, &prev);
 220         error = -ENOMEM;
 221         if (!vma)
 222                 goto out;
 223         if (unlikely(grows & PROT_GROWSDOWN)) {
 224                 if (vma->vm_start >= end)
 225                         goto out;
 226                 start = vma->vm_start;
 227                 error = -EINVAL;
 228                 if (!(vma->vm_flags & VM_GROWSDOWN))
 229                         goto out;
 230         }
 231         else {
 232                 if (vma->vm_start > start)
 233                         goto out;
 234                 if (unlikely(grows & PROT_GROWSUP)) {
 235                         end = vma->vm_end;
 236                         error = -EINVAL;
 237                         if (!(vma->vm_flags & VM_GROWSUP))
 238                                 goto out;
 239                 }
 240         }
 241         if (start > vma->vm_start)
 242                 prev = vma;
 243
 244         for (nstart = start ; ; ) {
 245                 unsigned long newflags;
 246
 247                 /* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
 248
 249                 newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
 250
 251                 /* newflags >> 4 shift VM_MAY% in place of VM_% */
 252                 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
 253                         error = -EACCES;
 254                         goto out;
 255                 }
 256
 257                 error = security_file_mprotect(vma, reqprot, prot);
 258                 if (error)
 259                         goto out;
 260
 261                 tmp = vma->vm_end;
 262                 if (tmp > end)
 263                         tmp = end;
 264                 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
 265                 if (error)
 266                         goto out;
 267                 nstart = tmp;
 268
 269                 if (nstart < prev->vm_end)
 270                         nstart = prev->vm_end;
 271                 if (nstart >= end)
 272                         goto out;
 273
 274                 vma = prev->vm_next;
 275                 if (!vma || vma->vm_start != nstart) {
 276                         error = -ENOMEM;
 277                         goto out;
 278                 }
 279         }
 280 out:
 281         up_write(&current->mm->mmap_sem);
 282         return error;
 283 }